MACHINE LEARNING LAB

UNIT I

1. What is Machine Learning?

 Answer: Machine Learning is a branch of artificial intelligence (AI) that involves the
development of algorithms that allow computers to learn from and make decisions
based on data. Instead of being explicitly programmed, the model learns patterns
from the data and improves its performance over time.

2. What are the types of Machine Learning?

 Answer: Machine Learning can be broadly classified into three types:

1. Supervised Learning: In this type, the model is trained using labeled data,
meaning that both input and output are provided. Examples include
classification and regression problems.
2. Unsupervised Learning: Here, the model is trained using data that has no
labels. It finds hidden patterns in the data. Examples include clustering and
association problems.
3. Reinforcement Learning: In reinforcement learning, the model learns
through trial and error by interacting with its environment and receiving
feedback in terms of rewards or punishments.

3. What is a well-posed learning problem in machine learning?

Answer: A well-posed learning problem has the following characteristics:

o A clear specification of the task to be learned.

o A set of training data that allows the learning process to be performed.
o A performance measure to evaluate the learned model’s success.

4. What are the key components required for designing a learning system?
Answer: The key components of a learning system include:

o Data: Labeled or unlabeled data that serves as input.

o Model: A hypothesis or function that learns from data.
o Learning algorithm: A procedure or algorithm for improving the model
based on data.
o Performance measure: A way to evaluate the model’s success.
o Environment: The external context or domain in which the learning system
operates.

5. What are the major challenges and issues faced in machine learning?
Answer: Some of the challenges in machine learning include:

o Data quality and quantity: Lack of sufficient or high-quality labeled data.

o Overfitting and underfitting: Balancing bias and variance in models.
o Computational cost: High computational cost for training large models.
o Interpretability: Difficulty in explaining how complex models make
decisions.
o Scalability: Handling large datasets and deploying models efficiently.

6.What is concept learning? How is it related to general-to-specific ordering?

Answer: Concept learning involves learning a definition or classification of a concept
from positive and negative examples. The general-to-specific ordering is a way of
organizing hypotheses, where general hypotheses are refined into more specific ones

Machine Learning Lab Viva Questions

 based on the examples provided. The goal is to identify the most specific concept that
satisfies the positive examples and rejects the negative ones.

7. Can you explain a concept learning task with an example?

Answer: A concept learning task involves defining a concept based on examples. For
example, learning the concept of "birds" from a set of positive and negative examples,
where positive examples are birds (e.g., "sparrow", "eagle") and negative examples
are non-birds (e.g., "cat", "fish"). The model learns the attributes that define a bird,
such as having wings and feathers.

8.What is the Find-S algorithm, and how does it work?

Answer: The Find-S algorithm is a simple algorithm for concept learning that finds the
maximally specific hypothesis that is consistent with the positive training examples. It
starts with the most specific hypothesis (e.g., a single example) and refines it by
generalizing based on each positive example in the training set. The final hypothesis is the
most specific hypothesis that fits all the positive examples.

9.What is a version space, and how is it used in machine learning?

Answer: A version space represents all hypotheses that are consistent with the training
examples. It is the set of all hypotheses that could explain the data, and the goal is to
narrow down the version space to find the most accurate hypothesis.
The candidate elimination algorithm systematically eliminates hypotheses from the
version space by using positive and negative examples. It maintains two sets: one for the
most specific hypothesis and one for the most general.

10.What are some challenges with version spaces and the candidate elimination
algorithm?
Answer: Some challenges include:

o Computational complexity: As the number of possible hypotheses grows,

the version space can become very large, making the search computationally
expensive.
o Noise in data: In real-world scenarios, noisy data can cause difficulties in
maintaining a clean version space, leading to incorrect eliminations.
o Scalability: The candidate elimination algorithm may not scale well with
large datasets or complex hypotheses.

11. What is inductive bias in machine learning?

Answer: Inductive bias refers to the assumptions made by a learning algorithm to
generalize from the training data to unseen examples. Without inductive bias, a
machine learning algorithm would be unable to learn effectively because it wouldn't
have any preference for which hypothesis is more likely. Examples of inductive bias
include assumptions about the simplicity of the model or the smoothness of the data.

12.What is Decision Tree Learning, and how does it work?

Answer: Decision tree learning is a supervised learning algorithm used for classification
and regression tasks. The model splits the dataset into subsets based on the feature
values, creating a tree-like structure where each internal node represents a feature and
each leaf node represents a class label or a value. The decision tree learns by recursively
partitioning the dataset, maximizing the homogeneity of the data in each split.

13.How is a decision tree represented?

Answer: A decision tree is represented as a tree structure with nodes and edges. Each
internal node corresponds to a feature (or attribute), and each branch represents a
possible outcome (decision) of that feature. Each leaf node corresponds to a predicted
class label or value (in regression problems). The tree is built by recursively splitting the
data based on the feature that best separates the data according to some criterion (such
as information gain or Gini index).
Machine Learning Lab Viva Questions
 14.What is hypothesis space search in decision tree learning?
Answer: Hypothesis space search in decision tree learning refers to the process of
searching through all possible decision trees to find the one that best fits the training
data. The search is guided by a criterion (e.g., information gain or Gini impurity) that
measures the quality of each potential split. The algorithm explores the space of all
possible trees by recursively choosing the best feature to split on and evaluating the
resulting subtrees until the tree structure is finalized.

15. What is the inductive bias in decision tree learning?

Answer: Inductive bias refers to the assumptions made by the learning algorithm that
guide it towards finding the best hypothesis. In decision tree learning, the inductive bias
includes:

o Preference for shorter trees: Decision tree algorithms tend to prefer

smaller, more specific trees that fit the data with fewer splits. This is why
pruning is often used.
o Greedy splitting: The algorithm makes decisions at each step based on
immediate gains (e.g., information gain or Gini index) without considering
long-term consequences.
o Feature importance: The decision tree algorithm assumes that features are
independent and can be used to split the data without considering
interactions between features.

16. .What is pruning in decision tree learning?

 Answer: Pruning is the process of removing branches from a decision tree after it
has been built to prevent overfitting. Pruning helps to simplify the model by
eliminating branches that provide little predictive value or that capture noise in the
data. There are two main types of pruning:
o Pre-pruning (early stopping): Stop growing the tree early by setting limits
on the depth or number of nodes.
o Post-pruning: Build the full tree first and then remove branches based on a
validation set or a complexity penalty.

Machine Learning Lab Viva Questions