AI, ML, DEEP

LEARNING AND
DATA MINING
METHODS FOR
HEALTHCARE
Knowledge discovery and Data Mining
◦ Traditionally, manual methods were employed to turn data into knowledge.
◦ However, sifting through huge amounts of data manually and making sense out of it is a slow, expensive,
subjective and prone to errors.
◦ Hence the need to automate the process arose; thereby leading to research in the fields of data mining and
knowledge discovery
◦ Knowledge discovery and data mining has recently emerged as an important research direction for
extracting useful information from vast repositories of data of various types.
◦ To store and retain immense amounts of data in easily accessible form comes with the challenge of
integrating it and make sense out of it.
◦ This raw data stores a huge amount of information, which, if utilized appropriately, can be converted into
knowledge, and would turn as boon for the human race.
◦ Hence..
◦ Knowledge discovery from databases (KDD) evolved as a research direction in
◦ databases,
◦ machine learning,
◦ pattern recognition,
◦ statistics,
◦ artificial intelligence,
◦ reasoning with uncertainty,
◦ expert systems,
◦ information retrieval,
◦ signal processing,
◦ high performance computing and
◦ networking.
Process of Knowledge Discovery
◦ The task of knowledge discovery can be classified into
◦ data preparation,
◦ data mining and
◦ knowledge presentation.
◦ In Data mining, algorithms extracts the useful and interesting patterns. the algorithms are
◦ C4. 5 for decision trees(generates decision, based on a certain sample of data),
◦ K-means for cluster data analysis(groups similar objects into clusters),
◦ Naive Bayes Algorithm(a supervised machine learning algorithm, for classification tasks, like text classification),
◦ Support Vector Mechanism Algorithms,
◦ The Apriori algorithm for time series data mining(the algorithm is used to find how two objects are associated and
related to each other).
◦ In this sense, data preparation and knowledge presentation can be considered, respectively, to be
preprocessing and post processing steps of data mining.
◦ Figure presents a process of knowledge discovery.
◦ Raw data from repository is send for preparation where it is collected and labeled.
◦ It is then cleaned (noise etc..), integrated and filtered that is preprocessed.
◦ Then the preprocessed data is mined based on the previous knowledge base.
◦ Patterns are extracted.
◦ The knowledge is presented to the user for visualization in terms of trees, tables, rules graphs, charts,
matrices, etc.
Data Warehousing
◦ An integral part of KDD is the database theory that provides the necessary tools to store, access and
manipulate data.
◦ In the data preparation step, the data is first cleaned to reduce noise and missing data.
◦ Once the data is cleaned, it may need to be integrated since there could be multiple sources of the data.
◦ After integration, further redundancy removal may need to be carried out.
◦ The cleaned and integrated data is stored in databases or data warehouses.
◦ A data warehouse includes:
◦ Cleaned and integrated data: This allows the miner to easily look across details of data without its
standardization, missing values and so on.
◦ Detailed and summarized data: Detailed data is necessary when the miner is interested in looking at the
data in its most granular form and is necessary for extracting important patterns.
◦ Summary data is important for a miner to learn about the patterns in the data that have already been extracted by
someone else.
◦ Summarized data ensures that the miner can build on the work of others rather than building everything from scratch.
◦ Historical data: This helps the miner in analyzing past trends/seasonal variations and gaining insights into
the current data.
◦ Metadata: This is used by the miner to describe the context and the meaning of the data.
◦ Data mining can be performed without the presence of a data warehouse, but data warehouses greatly
improve the efficiency of data mining.
◦ Since databases often constitute the repository of data that has to be mined, it is important to study how the
current database management system (DBMS) for efficient mining
◦ As a first step, it is necessary to develop efficient algorithms for implementing machine learning tools on top
of large databases and utilizing the existing DBMS support.
◦ The implementation of classification algorithms such as C4.5(Decision Tree Classifier) employed to generate
a decision, based on a certain sample of data ...
◦ or neural networks on top of a large database are required .
◦ In addition to this it is also necessary to develop new knowledge and data discovery management systems
(KDDMS) to manage KDD systems.
Data Mining
◦ Data mining is formally defined as the process of discovering interesting, previously unknown and potentially
useful patterns from large amounts of data.
◦ Patterns discovered could be of different types such as associations, subgraphs, changes, anomalies and
significant structures.
◦ Data mining techniques consist of three components:
◦ a model,
◦ a preference criterion and
◦ a search algorithm.
◦ The most common model functions in current data mining techniques include classification, clustering,
regression, sequence and link analysis and dependency modeling.
◦ Model must be flexible for representing the underlying data(info abt data source) and interpretability in
human terms.
◦ This includes decision trees and rules, linear and nonlinear models, example-based techniques such as
NN-rule and case-based reasoning, probabilistic graphical dependency models (e.g., Bayesian network) and
relational attribute models.
◦ The preference criterion is used to determine, depending on the underlying data set, which model to use
for mining.
◦ It tries to avoid overfitting of the underlying data .
◦ Finally, once the model and the preference criterion are selected, specification of the search algorithm is
defined in terms of these along with the given data.
Knowledge Presentation
◦ Presentation of the information extracted in the data mining in a format easily understood by the user is an
important issue in knowledge discovery.
◦ Important components of the knowledge presentation steps are
◦ data visualization and
◦ knowledge representation techniques.
◦ Presenting the information in a hierarchical manner is often very useful for the user to focus attention on
only the important and interesting concepts.
◦ This also enables the users to see the discovered patterns at multiple levels of abstraction.
◦ Some possible ways of knowledge presentation include:
◦ Rule(predefined linguistic rules to analyze and process textual data) and natural language generation(produces
human language text),
◦ tables and cross tabulations,
◦ graphical representation in the form of bar chart, pie chart and curves,
◦ data cube view representation, and
◦ decision trees
Tasks in Data Mining
◦ Data mining comprises the algorithms employed for extracting patterns from the data.
◦ Data mining tasks can be classified into two categories,
◦ descriptive and
◦ predictive
◦ The descriptive techniques provide a summary of the data and characterize its general properties.
◦ The predictive techniques learn from the current data in order to make predictions about the new data
sets.
◦ The commonly used tasks in data mining are described below.
Classification:

◦ Classification is a pattern recognition system that consists of three phases.

◦ These are
◦ data acquisition,
◦ feature extraction and
◦ classification.
◦ Data acquisition: depending on the environment within which the objects are to be classified, data are
gathered using a set of sensors, these are then passed on to the feature extraction phase,
◦ Feature extraction: where the dimensionality of the data is reduced by measuring/retaining only some
characteristic features or properties.
◦ Classification phase: The extracted features are passed on to the classifier that evaluates the incoming
information and makes a final decision.
◦ This phase basically establishes a transformation between the features and the classes.
◦ The problem of classification is basically one of partitioning the feature space into regions, one region for
each category of input.
◦ Thus it attempts to assign every data point in the entire feature space to one of the possible (say, k) classes.
◦ Classifiers are usually, but not always, designed with labeled data, in which case these problems are
sometimes referred to as supervised classification (where the parameters of a classifier function D are
learned).
◦ Some common examples of the supervised pattern classification techniques are the nearest neighbor (NN)
rule, Bayes maximum likelihood classifier and perceptron rule.
◦ Figure provides a block diagram showing the supervised classification process. Some of the related
classification techniques are described below.
◦ NN Rule:
◦ Some common examples of the supervised pattern classification techniques are the nearest neighbor (NN) rule
◦ The neighbor is nearby in feature space, it is likely to be similar to the object being classified and so is likely to be the
same class as that object.
◦ If the scheme employs the class label of only the nearest neighbor to pt.x, this is known as the 1-NN rule.
◦ If k neighbors are considered for classification, then the scheme is termed as the k-NN rule.
◦ Bayes Maximum Likelihood Classifier:
◦ Maximum likelihood estimation is a method of estimating the parameters of an assumed probability distribution, with
some observed data.
◦ This is achieved by maximizing a likelihood function so that, under the assumed statistical model, the observed data is
most probable
◦ using a probability model for data and optimizing the joint likelihood function of the observed data over one or more
parameters.
◦ Decision Trees:
◦ A decision tree is an acyclic graph, of which each internal node, branch and leaf node represents a test on a
feature.
◦ It is easy to convert any decision tree into classification rules.
◦ Once the training data points are available, a decision tree can be constructed from them from top to bottom
using a recursive divide and conquer algorithm.
◦ This process is also known as decision tree induction.
◦ A well known decision-tree induction algorithm, is described below.
◦ Decision tree induction (training data points, features)
1. Create a node N.
2. If all training data points belong to the same class (C) then return N as a leaf node labelled with class C.
3. If cardinality (features) is NULL then return N as a leaf node with the class label of the majority of the points
in the training data set.
4. Select a feature (F) corresponding to the highest information gain label node N with F.
5. For each known value fi of F, partition the data points as si.
6. Generate a branch from node N with the condition feature = fi.
7. If si is empty then attach a leaf labeled with the most common class in the data points.
8. Else attach the node returned by Decision tree induction(si,(features-F)).
◦ Regression:
◦ Regression is a technique used to learn the relationship between one or more independent (or, predictor) variables and
a dependent (or, criterion) variable.
◦ The simplest form of regression is linear regression where the relationship is modeled with a straight line learned using
the training data points.
◦ Cluster Analysis:
◦ When the only data available are unlabeled, the classification problems are referred to as unsupervised classification.
◦ Clustering is an important unsupervised classification technique where a set of patterns, usually vectors in a
multidimensional space, are grouped into clusters in such a way that patterns in the same cluster are similar in some
sense
◦ Patterns in different clusters are dissimilar in the same sense.
◦ For this it is necessary to first define a measure of similarity which will establish a rule for assigning patterns to a
particular cluster.
Major Issues and Challenges in Data Mining

◦ Major issues and challenges in data mining regarding underlying data types, mining techniques, user
interaction and performance are described
◦ Issues Related to the Underlying Data Types
◦ Complex and high dimensional data
◦ Databases with very large number of records having high dimensionality are quite common.
◦ These databases may contain complex data objects such as, hypertext and multimedia, graphical data,
transaction data, and spatial and temporal data.
◦ Mining these data may require exploring combinatorically explosive search space and may sometimes result in
spurious patterns.
◦ Therefore, it is important to develop a very efficient algorithms for data mining tasks which can also exploit
the advantages of techniques such as
◦ dimensionality reduction,
◦ sampling,
◦ approximation methods,
◦ incorporation of domain specific prior knowledge, etc.
◦ It is essential to develop different techniques for mining different databases, given the diversity of the data
types and the goals.
◦ For example,
◦ hybridization of several computational intelligence techniques for feature selection
◦ complex data that is modeled as a sequence of discrete multi-attribute records with two real applications, cleaning and
information extraction from websites,
◦ mining complex data represented as graphs
◦ tree mining.
◦ Missing, incomplete and noisy data
◦ Sometime data stored in a database either may not have a few important attributes or may have noisy values.
◦ These can result from operator error, actual system and measurement failure, or from a revision of the data
collection process.
◦ These incomplete or noisy objects may confuse the mining process causing the model to overfit/underfit the
data.
◦ As a result, the accuracy of the discovered patterns can be poor.
◦ Data cleaning techniques, more sophisticated statistical methods to identify hidden attributes and their
dependencies, as well as techniques for identifying outliers are therefore required.
◦ Handling changing data and knowledge
◦ Situations where the data set is changing rapidly (e.g., time series data or data obtained from sensors deployed
in real-life situations) may make previously discovered patterns invalid.
◦ Moreover, the variables measured in a given application database may be modified, deleted or augmented
with time.
◦ Incremental learning techniques are required to handle these types of data.
Issues Related to Data Mining Techniques
◦ Parallel and distributed algorithms
◦ The very large size of the underlying databases, the complex nature of the data and their distribution
motivated researchers to develop parallel and distributed data mining algorithms.
◦ Problem characteristics
◦ Though a number of data mining algorithms have been developed, there is none that is equally applicable to a
wide variety of data sets and can be called the universally best data mining technique.
◦ For example, there exist a number of classification algorithms such as decision-tree classifiers,
nearest-neighbor classifiers, neural networks, etc.
◦ When the data is high-dimensional with a mixture of continuous and categorical attributes,
decision-tree-based classifiers may be a good choice.
◦ However they may not be suitable when the true decision boundaries are nonlinear multivariate functions.
◦ In such cases, neural networks and probabilistic models may be a better choice.
◦ Thus, the particular data mining algorithm chosen is critically dependent on the problem domain.
Multi classifier Decision Fusion and
Ensemble
Learning
◦ The real-time reverse transcription-polymerize chain reaction (RT-PCR) is used to detect Coronavirus RNA
by using the mucus and saliva mixture samples taken by the nasopharyngeal swab technique.
◦ But, RT-PCR suffers from having low-sensitivity especially in the early stage.
◦ Therefore, the Chest radiography is used the early diagnosis of COVID-19 due to its fast imaging speed,
significantly low cost and low dosage exposure of radiation.
◦ Convolutional neural networks (CNNs) and ensemble learning idea is used in this case study to diagnose
X-ray images.
◦ Radiologists uses it as a supporting tool in COVID-19 detection.
◦ Deep feature are extracted by using seven CNN architectures were concatenated for feature level fusion and
fed to multiple classifiers in terms of decision level fusion idea with the aim of discriminating COVID-19,
pneumonia and no-finding classes.
◦ In the decision level fusion idea, a majority voting scheme was applied to the resultant decisions of classifiers
like
◦ softmax,
◦ RBF Kernel SVM(radial basis function kernel, is a popular kernel function used in various kernelized learning
algorithms) and
◦ Polynomial kernel SVM(a type of SVM that uses a polynomial function to transform the input data into a higher
dimensional space.)
◦ The obtained accuracy values were presented for three progressively created data sets.
◦ The proposed method results are superior to existing COVID-19 detection studies.
◦ The experimental results show that the proposed approach has attained high COVID-19 detection
performance that was proven by its comparable accuracy and superior precision/recall values with the
existing studies.
Meta-Learning
◦ Data about data is often called metadata
◦ Meta-learning refers to machine learning algorithms that learn from the output of other machine learning
algorithms.
◦ Meta-learning algorithms typically refer to ensemble learning algorithms like stacking that learn how to
combine the predictions from ensemble members.
◦ Meta-learning also refers to algorithms that learn how to learn across a suite of related prediction tasks,
referred to as multi-task learning.
◦ The approaches and applications in meta learning
◦ Meta learning is used in various areas of the machine learning domain.
◦ There are different approaches in meta learning as model-based, metrics-based, and optimization-based
approaches.
◦ Metric Learning:
◦ Means learning a metric space for predictions.
◦ This model gives good results in few-shot classification tasks.
◦ Model-Agnostic Meta-Learning (MAML):
◦ The neural network is trained by using a few examples to adapt the model to new tasks faster.
◦ MAML is a general optimization and task-agnostic algorithm, and it is used to train the parameters of a model for fast learning
with a small number of gradient updates.
◦ Recurrent Neural Networks (RNNs)
◦ Are a class/type of artificial neural networks, and they are applied to different machine learning problems, such as problems that
have sequential data or time series data.
◦ RNN models are commonly used for language translation, speech recognition, and handwriting recognition tasks.
◦ In meta learning, RNNs are used as an alternative to creating a recurrent model which can gather data sequentially from datasets
and process them as new inputs.
◦ Stacking/Stacked Generalization:
◦ Stacking is a subfield of ensemble learning and is used in meta learning models.
◦ Also, supervised learning and unsupervised learning models benefit from stacking.
◦ Stacking includes the following steps:
◦ Learning algorithms are trained by using available data
◦ A combiner algorithm (e.g., a meta learning model or a logistics regression model) is created to combine all predictions of these
learning algorithms, which refer as ensemble members.
◦ The combiner algorithm is used to make final predictions.
◦ What are the benefits of meta learning?
◦ Meta learning algorithms are used to improve machine learning solutions.
◦ The benefits of meta learning are higher model prediction accuracy:
◦ Optimizing learning algorithms: For example, optimizing hyperparameters to find best results. Thus, this optimization
task, which is normally done by a human, is done by a meta learning algorithm
◦ helping learning algorithms better adapt to changes in conditions
◦ identifying clues to design better learning algorithms a faster, cheaper training process
◦ Supporting learning from fewer examples
◦ Increasing the speed of learning processes by reducing necessary experiments
◦ Building more generalized models: learning to solve many tasks, not only one task: meta learning does not
focus on training one model on one specific dataset
Evolutionary algorithm
◦ An evolutionary algorithm is an evolutionary AI-based computer application that solves problems by
employing processes that mimic the behaviors of living things.
◦ Evolutionary algorithms function in a Darwinian-like natural selection process; the weakest solutions are
eliminated while stronger, more viable options are retained and re-evaluated in the next evolution—with the
goal being to arrive at optimal actions to achieve the desired outcomes.
◦ In simple words it is a stochastic search method that mimic the natural biological evolution.
◦ Evolutionary algorithms operate on a population of potential solutions applying the principle of survival of
the fittest to produce better and better approximations to a solution.
◦ At each generation, a new set of approximations is created by the process of selecting individuals according
to their level of fitness in the problem domain.
◦ This process leads to the evolution of populations of individuals that are better suited to their environment
than the individuals that they were created from, just as in natural adaptation.
◦ Evolutionary algorithms model natural processes,
such as selection, recombination, mutation
(changing structure of genes), migration, locality
and neighborhood.
◦ Figure shows the structure of a simple
evolutionary algorithm.
◦ The multi-population evolutionary algorithm models the evolution of a species in a way more similar to
nature than the single population evolutionary algorithm.
◦ Figure shows the structure of such an extended multi-population evolutionary algorithm.
◦ The evolutionary algorithms differ substantially from more traditional search and optimization methods.
◦ The most significant differences are:
◦ Evolutionary algorithms search a population of points in parallel, not just a single point.
◦ Evolutionary algorithms do not require derivative information or other auxiliary knowledge; only the
objective function and corresponding fitness levels influence the directions of search.
◦ Evolutionary algorithms use probabilistic transition rules, not deterministic ones.
◦ Evolutionary algorithms are generally more straightforward to apply, because no restrictions for the
definition of the objective function exist.
◦ Evolutionary algorithms can provide a number of potential solutions to a given problem.
◦ The final choice is left to the user.
◦ The following sections list some methods and operators of the main parts of Evolutionary Algorithms.
◦ Selection
◦ Selection determines, which individuals are chosen for recombination and how many offspring each selected
individual produces. The first step is fitness assignment by:
◦ proportional fitness assignment or
◦ rank-based fitness assignment,
◦ multi-objective ranking,
◦ The actual selection is performed in the next step. Parents are selected according to their fitness by means of
one of the following algorithms:
◦ roulette-wheel selection,
◦ stochastic universal sampling: a technique used in genetic algorithms for selecting
potentially useful solutions for recombination
◦ local selection: Local selection (LS) is a very simple selection scheme in evolutionary
algorithms. Individual fitnesses are compared to a fixed threshold, rather than to
each other, to decide who gets to reproduce
◦ Recombination
◦ Recombination produces new individuals in combining the information contained in the parents.
◦ The following algorithms can be applied:
◦ All presentation:
◦ discrete recombination: performs an exchange of variable values between the individuals. For each position the parent
who contributes its variable to the offspring is chosen randomly with equal probability.
◦ intermediate recombination: Is a method only applicable to real variables (and not binary variables). Here the variable
values of the offspring are chosen somewhere around and between the variable values of the parents.
◦ line recombination: produces new individuals in combining the information contained in two or
more parents
◦ extended line recombination:
◦ Binary valued recombination,
◦ single-point / double-point /multi-point crossover,
◦ uniform crossover,
◦ shuffle crossover,
◦ Mutation
◦ After recombination every offspring undergoes mutation.
◦ Offspring variables are mutated by small perturbations (size of the mutation step), with low probability.
◦ The representation of the variables determines the used algorithm.
◦ Two operators are explained:
◦ mutation operator for real valued variables,
◦ mutation for binary valued variables.
◦ Reinsertion
◦ After producing offspring they must be inserted into the population.
◦ Reinsertion determines which individuals should be inserted into the new population and which individuals
of the population will be replaced by offspring.
◦ The used selection algorithm determines the reinsertion scheme:
◦ global reinsertion for all population based selection algorithm (roulette-wheel selection, stochastic universal sampling,
truncation selection),
◦ local reinsertion for local selection.
◦ Population models - parallel implementation of evolutionary algorithms
◦ The extended management of populations (population models) allows the definition of extensions of
Evolutionary Algorithms.
◦ These extensions can contribute to an increased performance of Evolutionary Algorithms.
◦ The following extensions can be distinguished:
◦ global model,
◦ local model (diffusion model, neighborhood model, fine grained model),
◦ regional model (migration model, island model, coarse grained model).
◦ Competition between subpopulations
◦ Based on the regional population model the application of multiple different strategies at the same time is
possible.
◦ This is done by applying different operators and parameters for each subpopulation.
◦ For an efficient distribution of resources during an optimization competing subpopulations are used.
◦ application of multiple strategies,
◦ competition between subpopulations,
◦ These extensions of the regional population model contribute to an increased performance of Evolutionary
Algorithms, especially for large and complex real-world applications.
Evolutionary algorithm in home health care scheduling
◦ In many countries, home care clients are in need of assistance with basic activities of daily living due to temporary or long-term
physical or psychological disabilities.
◦ Consequently, the required care or support has a direct impact on the daily routines of the clients.
◦ The quality of care in a home care context largely depends on the coordination and timing of the service delivery.
◦ Problem description
◦ MILP formulation: In this section we formulate the problem as a Mixed-Integer Linear Programming (MILP) problem.
◦ Decision variables
◦ Formulation: the process of deciding what actions and states to consider, given a goal
◦ Permutation-assignment
◦ Encoding of solutions
◦ Fitness function
◦ Population
◦ Recombination of solutions
◦ Dependency measure
o Evolution of the population
◦ Termination condition
Computational Intelligence
◦ Computational Intelligence (CI) is the theory, design, application and development of biologically and
linguistically motivated computational paradigms.
◦ Traditionally the three main pillars of CI are
◦ Neural Networks,
◦ Fuzzy Systems and
◦ Evolutionary Computation.
◦ The CI is an evolving field and at present in addition to the three main constituents, it encompasses
computing paradigms like
◦ ambient intelligence: (AmI) is an intelligent computing where explicit input and output devices will not be required. a
sophisticated artificial intelligence system that recognizes and responds to human presence. Siri or Alexa are actual
examples of ambient intelligence
◦ artificial life: Artificial life (often abbreviated ALife or A-Life) is a field of study wherein researchers examine systems
related to natural life, its processes, and its evolution, through the use of simulations with computer models, robotics,
and biochemistry.
◦ cultural learning:
◦ artificial endocrine networks: The artificial endocrine system (AES) is a new branch of natural computing which uses
ideas and takes inspiration from the information processing mechanisms contained in the mammalian endocrine
system.
◦ it was thought that the major regulation systems of the human body—the nervous system, the immune system, and the endocrine
system—function independently.
◦ Now it is known that, all these systems are integrated into one single system of information communication. With cytokine,
neurotransmitter, and hormone, bidirectional information transmission among these regulation systems is achieved, and these
systems interact and cooperate with each other to organize a cubic and intelligent regulatory network.
◦ The enormous achievements made in artificial neural networks (ANNs) and artificial immune systems (AISs) have shown a clear
significance and value in both theory and application of intelligent computing and intelligent systems based on biological information
processing mechanisms.
◦ These results have inspired the research on other biological information processing systems, including the endocrine system.
◦ Comparatively speaking, research on artificial endocrine systems (AESs) is just at the stage of discipline creation
and preliminary exploration, and many challenges are yet to be discovered and overcome in theoretical models.
◦ Digital hormone model
◦ Computational model of hormones
◦ Artificial hormone system
◦ social reasoning: the ability to identify an appropriate action, given a work-related scenario.
◦ artificial hormone networks
◦ CI plays a major role in developing successful intelligent systems, including games and cognitive developmental systems.
◦ Over the last few years there has been an explosion of research on Deep Learning, in particular deep convolutional neural
networks.
◦ Nowadays, deep learning has become the core method for artificial intelligence. In fact, some of the most successful AI
systems are based on CI.
◦ Neural Networks
◦ Using the human brain as a source of inspiration, artificial neural networks (NNs) are massively parallel
distributed networks that have the ability to learn and generalize from examples.
◦ This area of research includes
◦ feedforward NNs,
◦ recurrent NNs,
◦ self-organizing NNs,
◦ deep learning,
◦ convolutional neural networks
◦ Fuzzy Systems
◦ Using the human language as a source of inspiration, fuzzy systems (FS) model linguistic imprecision and
solve uncertain problems based on a generalization of traditional logic.
◦ This area of research includes
◦ fuzzy sets and systems,
◦ fuzzy clustering and classification,
◦ fuzzy controllers,
◦ linguistic summarization,
◦ fuzzy neural networks,
◦ type 2 fuzzy sets and systems
◦ Evolutionary Computation
◦ Using the biological evolution as a source of inspiration, evolutionary computation (EC) solves optimization
problems by generating, evaluating and modifying a population of possible solutions.
◦ EC includes genetic algorithms, evolutionary programming, evolution strategies, genetic programming,
swarm intelligence, differential evolution, evolvable hardware, multi-objective optimization and so on.
Dimensionality Reduction
◦ Dimensionality reduction is an unsupervised learning technique.
◦ It is used as a data transform pre-processing step for machine learning algorithms on classification and
regression predictive modeling datasets with supervised learning algorithms.
◦ There are many dimensionality reduction algorithms to choose from and no single best algorithm for all
cases.
◦ Manifold Learning Methods
◦ Manifold learning methods seek a lower-dimensional projection of high dimensional input that captures the
salient properties of the input data.
◦ Some of the more popular methods include:
◦ Isomap Embedding
◦ Locally Linear Embedding
◦ Multidimensional Scaling
◦ Spectral Embedding
◦ t-distributed Stochastic Neighbor Embedding
◦ Each algorithm offers a different approach to the challenge of discovering natural relationships in data at
lower dimensions.
◦ There is no best dimensionality reduction algorithm, and no easy way to find the best algorithm for your data
without using controlled experiments.
Machine learning in healthcare
◦ Machine learning in healthcare is used to draw insights from large medical data sets to enhance clinicians'
decision-making, improve patient outcomes, automate healthcare professionals’ daily workflows, accelerate
medical research, and enhance operational efficiency.
◦ 10 benefits of machine learning in healthcare
1 Automated routine tasks From streamlining EHR processes to virtual nursing, machine learning can help healthcare professionals to automate many routine and
repeatable tasks.

2 Improved decision-making By detecting patterns in enormous healthcare data sets, machine learning helps clinicians to streamline diagnostics and improve
decision-making.

3 Enhanced patient experience With the help of chatbots and virtual assistants, healthcare organizations can improve patient experience by streamlining access to basic
healthcare services.

4 Accelerated innovation By optimizing drug discovery and development, machine learning helps pharmaceutical companies to reduce time-to-market and
dramatically decrease research costs.

5 Reduced operational costs Machine learning-enabled automation of back-office tasks allows healthcare organizations to cut costs and improve resource
management.

6 Improved patient outcomes With well-thought-out decision-making systems powered by machine learning, healthcare professionals can make more informed
decisions and improve patient outcomes.

7 Extended access to healthcare With efficiencies unlocked by machine learning, organizations can help more patients without compromising care quality.

8 Reduced risks Predictive analytics allows for early detection of serious illnesses, reduced risks during robot-assisted surgeries, and quick identification of
high-risk patients.

9 Increased staff satisfaction By allocating routine tasks to machine learning algorithms and enabling quick extraction of the most important data from patients’ cases,
healthcare organizations can substantially increase the satisfaction of their medical staff.

10 Refined data governance Machine learning algorithms can help healthcare organizations to consolidate disparate data sources and make sense of their large
datasets.
◦ Which machine learning algorithms are used the most in healthcare?
◦ 1. Artificial neural network
◦ Artificial neural network (ANN) is often referred to as the most ‘humanized’ machine learning algorithm.
ANNs sequentially filter incoming information based on set parameters and usually require minimum human
involvement during training.
◦ In the healthcare context, they are often used for medical imaging in radiology as well as text and speech
recognition.
◦ 2. Logistic regression
◦ Logistic regression is typically used to predict which outcome out of two is likely to happen.
◦ Its binary nature makes it comparatively easy to implement, which is why it’s one of the most popular
machine learning algorithms in healthcare.
◦ Besides predicting an outcome probability, logistic regression allows users to see how important each variable
is for the final outcome.
◦ Healthcare professionals use logistic regression for medical diagnosis, treating at-risk patients, and adjusting
behavior plans.
◦
◦ 3. Support vector machines
◦ Unlike linear regression algorithms, support vector machines (SVMs) are generally used for classification
problems.
◦ In simple terms, the further the data points are from the y axis on the graph below, the higher the probability
is that they belong to the respective classes.
◦ SVMs are frequently used to classify data from incomplete datasets with missing values and can be applied to
a range of healthcare tasks including drug development, medication adherence prediction, and image and text
segmentation.
◦
Machine learning in healthcare use cases
◦ 1. Patient behavior modification
◦ Many prevalent diseases are manageable or even avoidable. For example, type 2 diabetes, obesity, and heart diseases in
some cases can be avoided by practicing a healthier lifestyle.
◦ Adjusting the lifestyle requires a change in behavior, which is not a one-time effort but one requiring constant
reminders and follow-ups.
◦ For this purpose, machine learning algorithms can aggregate data incoming from patients' connected health devices
and sensors to generate insights into these patients' behavior and guide them during this transformational journey.
◦ Altering smoking habits with SmokeBeat
◦ SmokeBeat is an innovative application that passively gathers data on the user’s smoking behavior.
◦ The application uses an accelerometer on a smartwatch or a smart band to detect hand-to-mouth gestures. SmokeBeat
processes this data and offers real-time Cognitive Behavior Therapy incentives.
◦ User responses to those incentives are constantly measured and recorded to improve effectiveness.
◦ Additionally, SmokeBeat compares users' smoking data with their peers of choice, creating a sort of supportive social
network.
◦ 2. Virtual nursing
◦ In today’s busy hospital environment, nurses can become overworked and struggle to offer enough
personalized support to each patient.
◦ Healthcare facilities count on virtual nurses to solve this issue.
◦ Virtual nurses are computer-generated avatars that can interact with patients like humans.
◦ They are designed to be social, empathic, and informative.
◦ Virtual nurses can interact with patients more regularly than human nurses and answer questions in-between
doctor visits.
◦ They offer quick answers (faster than waiting for a nurse) and they are available 24/7.
◦ Streamlining remote care with Molly
◦ One example of a virtual nurse is Molly. This is a female avatar able to remotely monitor medical conditions,
which would be challenging to monitor on the spot. Molly receives data such as blood pressure and weight
from monitoring devices connected via Bluetooth. These devices are positioned in patients' homes, which
makes it convenient to take measurements as often as needed. Molly can recognize speech and verbally
answer patients' queries. Additionally, Molly understands human speech and can respond in kind. It also
offers a chatbot for private discussions.
◦ 3. Medical imaging
◦ Even with all the advancements in technology, medical image analysis is a tedious task prone to human error
since it requires great attention to detail.
◦ With the help of machine learning, it's possible to detect even the subtlest changes in medical scans.
Furthermore, traditional scan analyses (such as CAT scans and MRI) are time-consuming.
◦ Improving image quality with SubtleMR
◦ For example, SubtleMR developed by Subtle Medical is a machine learning-based software solution that
improves the quality of MRI protocols.
◦ With the help of denoising and resolution enhancement, SubtleMR can improve image quality and sharpness
with any MRI scanner and field strength.
◦ For example, RadNet, a US leader in outpatient imaging with 335 centers across the country, accelerated its
protocols by 33-45% after adopting SubtleMR technology.
◦ 4. Identifying high-risk patients
◦ By combining machine learning-powered pattern recognition and automation, clinicians can considerably
reduce the time it takes to identify high-risk patients.
◦ Detecting high-risk patients with UiPath and Amitech
◦ For example, UiPath, an AI- and RPA-enabled automation platform, was used by UiPath’s partner Amitech
to help a large healthcare provider automate patient analysis.
◦ A mixture of optical character recognition and natural language processing is used to structure and organize
patients’ medical documents.
◦ Then an RPA bot feeds this data to a machine learning system that scores patients for risks, provides them
with a tailored healthcare plan, and alerts the appropriate clinicians and care managers about high-risk
patients so the former can take immediate action.
◦ Besides significantly improving patient outcomes, instead of four weeks, the process now takes only minutes.
◦ 5. Robot-assisted surgery
◦ Using robots in healthcare is not a new trend. Robotic assistance in surgery increases precision, allows access
to different areas of the human body with minimal penetration and alleviates pressure from human surgeons
as robots can take over some parts of the work.
◦ Improving surgical outcomes with Senhance Surgical system
◦ Senhance Surgical system is a console-based, multiarmed surgical system that allows surgeons to remotely
control it.
◦ The system heavily relies on machine learning and deep learning models to bring the most challenging
healthcare ideas to reality.
◦ For example, during the preoperative stage, a machine learning-driven database allows surgeons to go
through simulation training.
◦ During surgeries, based on data from the eye-tracking camera, the system's Intelligent Surgical Unit can
automatically adjust the camera view and predict when a surgeon needs to zoom in or enhance images in
real-time.
◦ Drug discovery
◦ Drug discovery is an expensive and long process.
◦ Thousands of elements need to be tested together, and only one of them might end up as a viable drug.
◦ Machine learning algorithms are used in the drug discovery process for the following purposes:
◦ Minimizing clinical trial duration by predicting how potential drugs will perform
◦ Identifying combinations of existing drugs that can form a new treatment
◦ Discovering new drugs based on compound testing
◦ Finding new uses for previously tested substances
◦ Accelerating oncology research with IBM Watson
◦ Pfizer, a pharmaceutical company, is using IBM Watson for its immune-oncology research.
◦ While a human researcher can read around 300 articles a year, Watson was able to process one million journal
articles and data on four million patents.
◦ Given the ability of machine learning systems to ingest a huge amount of information, Pfizer employees can
identify non-obvious connections and help create treatment plans out of drug combinations.
◦ Finding cures to diseases with Google’s Deep Mind
◦ Another example is a machine learning-powered system called AlphaFold that can automatically predict
protein structure built by Google’s DeepMind and revealed in 2020.
◦ Reliably predicting how different proteins interact with each other is an immense biological breakthrough as
it can significantly accelerate drug screening and development.
◦ Last year, Google launched Isomorphic Labs, a company that will use AlphaFold’s technology to find cures
for prevalent diseases.
◦ 7. Hospital management optimization
◦ In general, far too many operations in healthcare institutions are undermined by ineffective management
practices.
◦ With the ever-growing demand for healthcare services, hospitals’ enterprise management systems are
becoming increasingly more chaotic.
◦ Carefully tuned machine learning-based systems can make sense of administrative data and handle the
majority of hospitals’ administrative functions.
◦ For example, Globus.ai is a Norway-based company that created a system to help healthcare institutions to
streamline staffing.
◦ With the help of natural language processing and machine learning, the system can match healthcare
employees to specific tasks based on their skill sets, making task scheduling far more efficient. Importantly,
Globus.ai’s system considers legal requirements when making scheduling decisions.
◦ For example, in some cases, the law limits the number of working hours or requires a professional with
particular expertise to be present during a certain procedure.
◦ 8. Disease outbreak prediction
◦ Nowadays, a huge amount of data can be collected from satellites. This includes real-time data from social
media and other historical web data.
◦ Machine learning algorithms help aggregate this data and make predictions about potential disease outbreaks.
◦ One example is predicting malaria outbreaks by analyzing data including monthly rainfall, temperature, and
similar parameters.
◦ This can be particularly relevant in third-world countries that lack medical infrastructure and the necessary
education to combat those diseases.
◦ Knowing about such critical outbreaks upfront will allow precautions to be taken in order to minimize their
negative impact and save lives.
◦ ProMED reporting disease outbreaks online
◦ ProMED (the Program for Monitoring Emerging Diseases) offers an online real-time reporting system
showing outbreaks of infectious diseases worldwide and any exposure to toxins affecting human or animal
health. ProMED aggregates data from sources such as official reports, media reports, local observers, and
reports contributed by its subscribers. An expert team reviews these reports before they are accepted into the
system.
◦ The data provided by ProMED is aggregated by HealthMap to visualize disease outbreaks in every country.
◦ 9. Medical diagnostics
◦ In healthcare, inaccurate or incomplete diagnosis of diseases can be detrimental to patient outcomes, and, in
the worst-case scenarios, lead to death.
◦ To address one of the most apparent healthcare challenges, many companies are tapping into machine
learning to make medical diagnostics more accurate.
◦ Predicting syndromes with Face2Gene
◦ A great example is the Face2Gene app, a machine learning-enabled facial recognition software that helps
clinicians to more accurately diagnose rare diseases.
◦ With the help of machine learning, Face2Gene can detect phenotypes, reveal relevant facial features, and
evaluate the probability of a patient having a particular syndrome.
◦ 10. Health insurance
◦ Health insurance is an essential component of the healthcare industry and plays a critical role in enhancing
access to healthcare. However, there is plenty of room for improvement in the mostly manual-based
processes of health insurance.
◦ For example, machine learning-enabled pattern recognition algorithms can assist in early fraud detection.
Rule-based fraud detection systems that the majority of health insurers currently use can flag too many claims
as potentially fraudulent. Machine learning systems, on the other hand, learn and gradually decrease the
probability of false positives.
◦ Machine learning can also be useful for the automation of different health insurance processes, including
credit underwriting, risk assessment, claims to process, and customer support.
◦ Optimizing health insurance with Maya Intelligence
◦ For example, Temple University Health System (TUHS), a nationally-recognized academic health system in
Philadelphia partnered with Accolade, a company that provides the Maya Intelligence platform to help
patients choose the most appropriate healthcare coverage option. The system utilizes machine learning to
analyze medical claims, lab results, and other relevant patient information to offer tailored healthcare plans to
patients. As a result of the implementation, TUHS has managed to save more than $2 million in healthcare
claim costs, and achieve a 50% increase in employee engagement.
 Machine learning challenges in healthcare
While machine learning has immense potential to transform healthcare, it’s critical to consider the challenges and risks associated with its implementation.
1. Lack of data 2. Bias
The lack of clean, structured data is an overarching problem for organizations across While all machine learning applications can suffer from bias, its implications in
every industry, but training and deploying value-adding machine learning models at healthcare are justifiably the most concerning. To put it simply, given that it's
scale requires companies to reimagine their approaches to data governance. Given humans who train machine learning algorithms, our existing biases inevitably creep
that datasets from one organization rarely suffice for model training, engineers in. What’s even more daunting is that ML models do not only sustain these
typically resort to obtaining patient data from other healthcare organizations. The prejudices but often amplify them.
problem is that the majority of these datasets are incompatible with each other. Fortunately, the importance of unbiased and equitable data is getting recognized at
Enforcement of industry-wide data governance frameworks is paramount at this the governmental level. In 2021, the Biden administration formed the Equitable
point. Standardization of medical data requires equal effort from both governmental Data Working Group to ensure that historically underserved communities get equal
bodies and industry players. access to healthcare services. Importantly, enforcing robust standards to make
Thankfully, such initiatives are already taking place. In 2022, the White House-led different datasets interoperable is also on the group’s agenda. Both NAIRR’s and
National AI Research Resource (NAIRR) team issued a comprehensive interim Equitable Working Group’s initiatives are instrumental in maximizing the
report that outlines recommendations around data collection and aims to extend effectiveness of machine learning and reducing the probability of bias.
access to medical data for more AI companies and researchers.
3. Lack of strategy 4. Lack of in-house expertise
In the absolute majority of cases, machine learning brings tangible long-term benefits Integrating such a complex technology as machine learning into intricate healthcare
when all parts of the organization support its adoption. Given that machine learning workflows requires both excellent technical skills and a deep understanding of
has a much more drastic impact on conventional healthcare workflows than the medical science. On the one hand, many ambitious AI startups fail to incorporate
majority of other technologies, companies should make an effort to redefine team clinical expertise during the early phases of development, while, on the other hand,
roles, invest in change management, and launch workforce reskilling programs. many credible and experienced clinicians have insufficient understanding of machine
Understandably, many organizations are reluctant to undergo such significant learning to provide tangible input.
changes for the sake of technology adoption. However, according to a recent This is why for artificial intelligence startups tapping into healthcare, assembling a
Deloitte survey, the majority of industry leaders believe that these initiatives will multifaceted team is a prerequisite for success. It’s not only important to hire
inevitably prove worthwhile in the long term. exceptional talent in multiple fields but to also ensure that data scientists, machine
learning engineers, medical professionals, legal advisers, and other experts can
collaborate with each other.
Deep Learning in Healthcare
◦ The ML can be solving multiple tasks but with human interventions.
◦ While the DL performs alone.
◦ Deep learning solves the whole problem unlike ML automatically.
◦ Deep learning is more beneficial for elders, coma patients, and cardiac disease diagnosis, especially in case of
children.
◦ Deep learning is beneficial for the recognition of any problem, and the data-driven performance is also better
in any type of problem.
◦ The learning algorithms are simple, i.e., easily learned or performed and easily able for application.
◦ Recent deep learning models for healthcare applications, on the other hand, are sophisticated and need a
great amount of computing resources for training and prediction.
◦ It also takes a long time to train and evaluate data using these sophisticated neural networks.
◦ The more advanced the network and the longer the forecast period, the higher the accuracy required.
◦ This has been a significant issue in healthcare and other IoT applications where real-time findings are crucial.
◦ The deep learning is gifted to deliver the potential for enhancements in worth control engineering system.
◦ Machine learning is responsible to provide beneficial tools for understanding the domain.