Automatic Detection of Liver Diseases based on

Supervised Learning Algorithms

Abdelmoty M. Ahmed

Nahda University
Belgacem Bouallegue
King Khalid University
Mahmoud M. Khattab
International Islamic University Malaysia
Hegazi Ibrahim
Nile Higher Institute for Engineering and Technology
Ahmed G. Mabrouk
High Institute of Computers and Information

Research Article

Keywords: fatty liver, feature extraction, learning algorithms, classifier, resampling, ultrasound

Posted Date: April 29th, 2024

DOI: https://doi.org/10.21203/rs.3.rs-4298719/v1

License:   This work is licensed under a Creative Commons Attribution 4.0 International License.
Read Full License

Additional Declarations: No competing interests reported.

Page 1/20
Abstract
The incidence of Liver disease has been steadily rising due to the excessive consumption of alcohol,
exposure to harmful gases through inhalation, consumption of contaminated food, and the use of certain
drugs. Dataset was used that extracted from ultrasound images for liver, and some chemical compounds
(bilirubin, albumin, proteins, alkaline phosphatase) present in human body as features, to build and
evaluate a classifier model, it can classify and differentiate liver diseases. This paper with the help of
machine learning technique proposes diagnosing and classifies liver diseases into liver patient
(abnormal) and non-liver patient (normal), then classify the liver patient into fatty liver or cirrhosis, using
many supervised learning algorithms. The findings indicated that the Random Forest (RF) algorithm
exhibited superior accuracy, achieving an impressive 95.12% accuracy rate. Following closely was the
Decision Tree (DT) classifier, achieving a respectable accuracy of 90.24%. These results were obtained
after employing a data reduction technique involving resampling. The model's performance was
subsequently assessed using a 10-fold cross-validation (CV) approach, widely regarded as the optimal
method for classifier evaluation. This approach leverages resampling across various folds of the dataset
during multiple iterations, enhancing the classifier's ability to generalize and consequently yielding
elevated accuracy when applied to unlabeled image samples.

1. INTRODUCTION
Diseases affecting the liver, such as fatty liver and cirrhosis, which manifest in a diffuse manner, stand as
prominent contributors to disability and mortality on a global scale. Timely identification of these
conditions holds paramount significance in terms of preserving lives and enhancing the efficacy of
medical interventions. Liver ailments hold a significant position among the most widespread health
concerns worldwide, presenting a considerable threat to society's well-being [1][2]. Recently, the use of
artificial intelligence (machine learning- supervised learning) has spread in the medical sector due to the
importance of community health. In this study proposes a non-invasive method for diagnosing liver
diseases based on ultrasound images and some chemical compounds such as bilirubin, albumin,
proteins, alkaline phosphatase present in human body, based on machine learning algorithms to classify
the liver tissue into normal, or abnormal, based on a set of techniques mainly: feature extraction [3][4],
feature selection to identify the most significant features in training dataset using Information gain and
Gini index, then features resampling, it is the creation of new samples based on one observed sample.
Our classification method provides two key contributions: combination of two different feature selection
methods. Many feature are selected to learn the model such as the correlation, homogeneity, variance,
entropy, contrast, energy, long run emphasis, run percentage, and standard deviation are determined. After
that apply the classification algorithm using supervised learning mainly: Multinomial NB, Gaussian NB,
Bernoulli NB, SVC, Linear SVC, and NuSVC, etc. Finally model evaluation and its testing, this model is
tested using cross-validation method, it is a statistical method for validating a predictive model, n folds of
the data are held out for use as validating sets; each fold of them is held out once to test the model and
the classifier is trained on the remaining (n-1) [5]. ​Averaging the quality of the predictions across the

Page 2/20
validation sets yields an overall measure of prediction accuracy. Cross-validation is employed repeatedly
in building many algorithms such as decision trees, naive Bayes, and random forest, etc.

In [6], the authors developed a Computer-Aided Diagnosis technique using machine learning algorithms
and a voting-based classifier to categorize liver tissues as either fatty or normal. They extracted
ultrasound image features and achieved high accuracy in classifying liver images without a
segmentation phase. The proposed method achieved a classification accuracy of 95.71% using the
voting-based classifier and 93.12% using the J48 algorithm.

In [7], several classifiers were compared, including machine learning and deep learning methods, for
binary classification of liver disease. In [8], authors evaluate various machine learning models for
predicting liver disease risk. The Voting classifier outperformed other models with an accuracy of 80.1%,
recall of 80.4%, and F1-score of 88.4% after SMOTE with 10-fold cross-validation.

This study aims to build a model that is able to distinguish and classify liver diseases into normal and
abnormal, then predict the abnormal liver into fatty or cirrhosis liver, then determining the best way to
treatment liver disease patients to help the medical sector in diagnosing and predicting disease, providing
an integrated health system for liver disease patients to reduce or eliminating infection with liver disease
through accurate diagnosis, providing appropriate treatment, improving response to treatment and its
effectiveness, and predicting the disease at the level of the next generation.

2. ACCURACY OF SUPERVISED LEARNING ALGORITHMS FOR LIVER

DISEASE DETECTION
Improving the accuracy of supervised learning algorithms for liver disease detection can be approached
from several angles as follow (i) Data Quality: Ensuring high-quality, diverse, and representative data is
crucial. This includes careful preprocessing to handle missing values, noise, and outliers; (ii) Feature
Engineering: Selecting the most relevant features or creating new features that capture the underlying
patterns in the data can significantly enhance model performance [8], (iii) Advanced Algorithms:
Exploring more sophisticated machine learning models like deep learning (e.g., Convolutional Neural
Networks (CNNs) combined with Long Short-Term Memory networks (LSTMs)) can lead to better
accuracy [8]; (iv) Ensemble Methods: Combining predictions from multiple models (e.g., using a Voting
classifier) can improve accuracy by leveraging the strengths of individual models [9]; (v) Cross-validation:
Using techniques like k-fold cross-validation helps in assessing the model's performance more reliably
and ensures that it generalizes well to unseen data; (vi) Regularization: Implementing regularization
methods to prevent overfitting, which can improve the model's performance on new, unseen data; (vii)
Hyperparameter Tuning: Systematically searching for the optimal hyperparameters can fine-tune the
model's ability to learn from the data; and (viii) Clinical Validation: Collaborating with healthcare
professionals to validate the models in a clinical setting can provide feedback for further refinement. By
focusing on these areas, the accuracy and reliability of liver disease detection algorithms can be
enhanced, leading to better diagnostic tools in healthcare [7–9].

Page 3/20
The accuracy of supervised learning algorithms in the detection of liver diseases can vary depending on
the specific algorithm used, the quality of the data, and the complexity of the task. However, recent
studies have shown promising results:

1. One study reported that a Voting classifier achieved an accuracy of 80.1%, a precision of 80.4%, and
an area under the curve (AUC) of 88.4% after Synthetic Minority Over-sampling Technique (SMOTE)
with 10-fold cross-validation [8].
2. Another research demonstrated a deep learning model for detecting liver disease tumors using CT
images, which achieved a dice similarity coefficient value of 98.59%, indicating high accuracy in
classification [7].
3. A decision tree model was proposed that accurately classified 93.7% of patients [10].

The automatic detection of liver diseases using supervised learning algorithms involves high level
processes as shown in Fig. 1 [8, 11]. The process of diseases detection controlled by several concepts as
: (i) Gathering a dataset of liver disease cases, including both healthy and diseased samples, (ii) Clean
the data to handle missing values, normalize or standardize it, and possibly perform feature selection or
extraction, (iii) Choose appropriate supervised learning algorithms (e.g., SVM, Naïve Bayes, K-NN, CART,
LDA), (iv) Train the selected algorithms on a portion of the dataset to learn the patterns associated with
liver diseases, (v) Validate the models using a separate portion of the dataset to asse their performance
(accuracy, precision, recall, F-measure, AUC), (vi) Fine-tune the models by adjusting hyperparameters and
using techniques like cross-validation, (vii) Implement the best-performing model into a clinical setting for
real-world liver disease detection, and (viii) Continuously monitor the model's performance and update it
as necessary to maintain accuracy over time [11].

3. RESEARCH METHODOLOGY
The methodology of this study is based on liver diseases classifications according to mix of the features
that are extracted from ultrasound images and chemical compounds such as bilirubin, albumin, proteins,
alkaline phosphatase; starts with collect images that are labeled into predefined classes normal and
abnormal (fatty and cirrhosis) as shown in Fig. 2, This data set contains two different forms, the first
dataset that have extracted from ultrasound images [12], it consists of 123 samples of liver divided as 41
for Normal, 41 for Fatty and 41 for Cirrhosis. The second dataset is chemical compounds such as
bilirubin, albumin, proteins, alkaline phosphatase, it consists of 583 samples of liver divided as 416 for
abnormal, and 167 for normal that are labeled into predefined classes normal and abnormal (fatty and
cirrhosis), then feature selection, learning algorithms, and evaluate the classifier.

Figure 2 Samples of the acquired images of normal, fatty and cirrhosis liver

At the onset of our investigation, we embarked on the pivotal task of feature selection—an endeavor
geared toward identifying the most pertinent attributes that wield the power to effectively distinguish
between classes within the dataset. This approach stands as an adept solution in resolving the
complexities inherent to high-dimensional data scenarios. Our toolkit comprises a trio of robust statistical
Page 4/20
techniques: Information Gain, Chi-Square, and Gain Ratio methodologies [13]. These techniques,
grounded in sound statistical principles, play a crucial role in quantifying the significance of attributes.
Illustrated in Fig. 3 and Fig. 4, we present a visual representation of this intricate process, spotlighting the
attributes that rise to prominence with the highest rankings for each respective technique. Notably, the
gain ratio, defined by Eq. 4, stems from the calculation of attribute gain via Eq. 3. The attribute boasting
the utmost gain ratio is judiciously chosen as the pivotal splitting attribute, facilitating streamlined
classification. Similarly, the Chi-Square technique, expounded upon by Eq. 2, stands as an essential
contributor in the attribute selection process. Finally, the Information Gain, framed within Eq. 1
n
nfo (D) = −∑ Pi log (Pi )
2
i=1

(Observedvalues − Expectedvalues) 2 (O i− Ei ) 2
2
X = ∑ = ∑
C
Expectedvalues Ei

2
The term XC
2
represents the squared difference between the observed (O) and expected (E) values,
divided by the expected value within one or more classes (C).

The gain ratio serves as an enhancement of the information gain, specifically designed to mitigate its
inclination towards attributes with extensive branching. Computed as per Eq. 3, the gain ratio offers a
refined perspective. Ultimately, the attribute exhibiting the highest gain ratio is singled out as the optimal
candidate for splitting, following the principles outlined in Eq. 4

n |Dj| |Dj|
SplitInfoA (D) = −∑j=1 x log2 ( ) (3)
|D| |D|

Gain(A)
Gain Ratio (A) = (4)
SplitI nf o(A)

Learning algorithms are engineered with the objective of constructing classification models or training
the algorithm itself to discern patterns and crucial characteristics for each category. This process unfolds
through utilizing a training dataset, allowing the algorithm to form predictions regarding the category of
liver disease. Various algorithms have been harnessed within this study to create a model capable of
effectively categorizing liver conditions. This involves training the model using designated training data
and subsequently evaluating its performance on unseen test data to ascertain its accuracy. The process
initiates by partitioning the dataset into training and testing subsets, a procedure facilitated through the
cross-validation method [8]. This comprehensive approach ensures a robust evaluation of the model's
predictive capabilities and underscores its aptitude for accurate liver category classification, then features
selection to learn the model as shown in Fig. 5. Many algorithms are used in this study mainly:

Page 5/20
The assessment of classifier accuracy is a comprehensive endeavor that hinges on a variety of metrics,
with primary emphasis placed on Recall, Precision, and the F1-score. This evaluation unfolds after the
computation of critical factors including the true positives rate (TP), true negatives rate (TN), false
negative rate (FN), and false positive rate (FP) [15].

To elaborate, TP and TN correspond to the count of samples that have been correctly assigned to their
designated category. In contrast, FP and FN represent the instances in which samples have been
inaccurately excluded from or attributed to the category in question. These foundational metrics
contribute to a nuanced understanding of the classifier's performance, ultimately shaping our perception
of its efficacy in accurately classifying liver conditions.

TP: The number of samples which are correctly assigned to the given category, “sick people correctly
identified as sick”
TN: The number of samples which are correctly assigned not to belong to category, “Healthy people
correctly identified as healthy”
FP: The number of samples which are incorrectly assigned to the category, “Healthy people
incorrectly identified as sick”.
FN: The number of samples which are incorrectly not assigned to the category, “Sick people
incorrectly identified as healthy”.

Accuracy checks how many samples are classified correctly. It provides an evaluation of how much the
outcomes are matched to the original outcome according to Eq. (5)

TP + TN
Acc =
TP + TN + FP + FN

5
Precision estimates how precise the behavior of the suggested method is, by analyzing the correct TPs
from the forecasted ones. according to Eq. (6)

TP
P recision =
TP + FP

6
Recall or Sensitivity is defined as the rate of correct detection of samples that are positives. It provides
the possibility of attaining the correct test outcome. As in Eq. (7)

TP
Sensitivity =
TP + FN

7
The F-measure serves as the harmonic means between precision and recall, affording equal importance
to both metrics. This measure provides a unified assessment that encapsulates both precision and recall,
Page 6/20
utilizing a singular score. This is particularly valuable when evaluating a model's performance and
making comparisons between different models. according to Eq. (8)

2 (precision ∗ recall)
F − measure =
precision + recall

4. MACHINE LEARNING CLASSIFIERS

There are different machine learning classifiers used in this paper as discussed in table 1 to optimize the
performance by improving the accuracy based on the problems, available data, and application
requirements [17–20].

Page 7/20
 Table.1 Machine Learning Classifiers

Classifier Meaning and Benefits

Multinomial Multinomial Naive Bayes is a probabilistic classifier that calculates the probability
Naive distribution of text data. It is well-suited for data with features representing discrete
Bayes frequencies or counts of events, commonly used in natural language processing
(M_NB) (NLP) tasks. It assumes independence between features, making it efficient for large
vocabularies [17].

Bernoulli Bernoulli Naive Bayes is used for binary classification tasks (e.g., spam detection,
Naive sentiment analysis). It predicts the probability of a binary outcome (e.g., presence or
Bayes absence of a certain word) using the Bernoulli distribution. Features are treated as
(B_NB) binary (present/absent) and assumed to be independent [18].

Logistic Logistic Regression predicts the probability of a binary outcome based on a linear
Regression combination of features. It uses the logistic function (sigmoid) to map linear
(LR) combinations to a probability range (0 to 1). Commonly used for binary classification
and probability estimation [17].

Stochastic SGD updates model parameters iteratively using mini batches of training data. It
Gradient minimizes a loss function by adjusting parameters in the direction of the negative
Descent gradient. Efficient for large datasets and real-time learning [19].
(SGD)
Classifier

Support SVC seeks an optimal hyperplane to separate two classes in a dataset. The decision
Vector function is based on a linear combination of features. Works well for both linear and
Classifier non-linear separations [19].
(SVC)

Linear Like SVC but specifically designed for linear kernels and it’s useful for linearly
Support separable data [19].
Vector
Classifier
(L_SVC)

Nu-Support Like SVC but with different mathematical formulations. Provides flexibility in handling
Vector non-linear data [19].
Classifier
(Nu-SVC)

Decision Decision trees recursively split data based on feature thresholds to create a tree-like
Tree structure. Each node represents a decision rule based on a feature. Final prediction is
Classifier based on the majority class in the leaf node [20].
(DT)

Random Random Forest combines multiple decision trees to improve accuracy and reduce
Forest overfitting. Aggregates predictions from individual trees. Widely used for classification
Classifier and regression tasks [20].
(RF)

5. PROPOSED SYSTEM
The proposed system encompasses a systematic framework, comprising four distinct stages, each
crucial to achieving accurate liver disease classification. The initial stage involves feature extraction from
Page 8/20
images, coupled with the selection of paramount features pertinent to the final class. This critical feature
selection process employs techniques like Information Gain and Gini Index.

Following this, the dataset undergoes partitioning into training and testing subsets, orchestrated through
the cross-validation method. This serves as the second stage, facilitating a robust evaluation of the
system's predictive capabilities.

In the third stage, the crux of supervised learning is employed. Utilizing a collection of pre-classified
samples pertinent to liver disease classification, the learning algorithms are fine-tuned. This stage
generates a classifier model that holds the ability to distinguish between various liver conditions.

The final stage culminates with a comprehensive evaluation of the classifier's performance. Utilizing
unseen instances of liver disease, the classifier's predictive prowess is tested, ultimately predicting the
appropriate liver disease class.

This intricate interplay of stages is adeptly illustrated in Fig. 6, collectively forming a holistic and efficient
system for accurate liver disease classification.

6. EXPERIMENTAL RESULTS
In this phase, an array of classifiers was meticulously employed within the study to ascertain the optimal
classifier that exhibits the highest accuracy in predicting liver disease. The evaluation process was
conducted under two distinct scenarios: prior to and after the implementation of resampling. Resampling,
which involves generating random subsamples of a dataset through techniques like sampling with or
without replacement, was applied to enhance classifier performance.

Initially, the average accuracy of these classifiers stood at 71.01% prior to resampling. However, upon
integrating resampling techniques, this figure significantly increased to 89.02%, indicative of its efficacy
in improving predictive accuracy. Furthermore, a similar trend was observed in a separate evaluation,
where the accuracy initially measured at 80.48% surged to an impressive 95.12% after the application of
resampling techniques. This augmentation in accuracy was distinctly witnessed across two distinct
datasets - ultrasound and chemical compounds - as visually depicted in Tables 2 and 3, also shown in
Figs. 7 and 8.

It is noteworthy that these experimental results underwent rigorous assessment and validation using both
Python and WEKA programs, further affirming the credibility and robustness of the findings.

Page 9/20
 Table 2
Performance of Different Leading Classifiers
Dataset Dataset-1 Dataset-2

Classifier (Ultrasound Images) (Chemical Compounds)

Before Sampling After Sampling Before Sampling After Sampling

RF 71.01 89.02 80.48 95.12

DT 69.12 83.53 68.29 90.24

Assessing the efficacy of the random forest and decision tree classifiers necessitates a comprehensive
evaluation that involves understanding the intricacies of the confusion matrix [16]. This matrix enables
the determination of properly classified and misclassified samples by revealing the count of True
Positives (TP), True Negatives (TN), False Positives (FP), and False Negatives (FN) for each class. Each
row within the matrix signifies the number of instances predicted as belonging to a particular class, while
each column indicates the count of instances in the corresponding actual class.

The performance evaluation of these classifiers encompasses several crucial metrics, primarily focusing
on precision, recall, and the f1-score (f-measure), as computed through equations 5, 6, 7, and 8. These
measures provide an insightful analysis of the classifiers' ability to accurately identify and categorize
instances, offering a comprehensive understanding of their classification prowess and predictive
capabilities.

Table 3
Performance Metrics for Different Classifiers
Class TP Rate FP Rate Precision Recall F1-score ROC Area

RF Abnormal 0.96 0.246 0.98 0.96 0.99 0.98

Normal 0.99 0.055 0.95 0.99 0.92 0.95

avg / total 0.97 0.191 0.97 0.97 0.97 0.97

DT Abnormal 0.95 0.246 0.95 0.95 0.95 0.95

Normal 0.91 0.055 0.91 0.91 0.91 0.91

avg / total 0.94 0.191 0.94 0.94 0.94 0.94

Figure 8 depicts the True Positive (TP) rate, False Positive (FP) rate, F1-score, Recall, and Precision values
corresponding to each category. These metrics were evaluated for both the Random Forest (RF) and
Decision Tree (DT) classifiers, employing the Cross-Validation (CV) method.

Continuing with the analysis, the visual representation in Fig. 7 elucidates the comprehensive
performance assessment of the Random Forest (RF) and Decision Tree (DT) classifiers using the Cross-
Validation (CV) method. Each category's True Positive (TP) rate, False Positive (FP) rate, F1-score, Recall,
Page 10/20
and Precision metrics are meticulously showcased, providing a comprehensive insight into the classifiers'
capabilities across various classification aspects. This graphical depiction offers a succinct yet detailed
overview of how RF and DT classifiers perform within the context of each category, aiding in a deeper
understanding of their strengths and nuances in classification accuracy and predictive power.

Also, according to the observations based on the available data, DT and Random RF models have perfect
training accuracy (1.0), but their testing accuracy is also high, indicating good generalization, however
SVC has high training accuracy but lower testing accuracy, suggesting potential overfitting, finally B_NB,
LR models have similar training and testing accuracies, and SGD model has relatively lower testing
accuracy as shown in Fig. 9.

Consequently, available data for different classifiers and their training accuracies have been analyzed
and compared as discussed in table 4, also they have been mentioned in Figs. 10 and 11 for the training
and prediction ones as follow:

Table 4. Analysis of Classifier’s Performance Accuracies (Training Phase)

Classifier Training Predicting Observations

Accuracy Accuracy

M_NB 40% 35% I. Models like SVC, NuSVC, DT, and RF achieved perfect
training accuracy, which might be a sign of overfitting.
B_NB 59% 67%
II. B_NB and SGD had relatively better training accuracies.
LR 40% 67%
III. LR, M_NB, and L_SVC performed poorly during training.
SGD 59% 49%
IV. DT and RF models seem to be overfitting the training
SVC 100% 67% data due to their high accuracy.

L_SVC 40% 60% V. SGD has the lowest training accuracy’s, LR, and B_NB
show consistent performance during training.
NuSVC 100% 67%

DT 100% 95%

RF 100% 96%

7. CONCLUSIONS
Artificial intelligence (AI) systems have gained significant traction within the realm of medical
applications, playing a pivotal role in both diagnosis and treatment processes. These systems act as
invaluable aids to physicians, assisting them in arriving at informed decisions during the diagnostic
phase. Furthermore, AI is harnessed to anticipate diseases prior to their onset or during their nascent
stages through the utilization of advanced learning algorithms. Within this study, a diverse array of
classifier models was meticulously constructed through the utilization of multiple algorithms. Among
these models, we have highlighted the performance of the top two classifiers. The empirical findings
Page 11/20
underscore the supremacy of the Random Forest (RF) classifier, exhibiting an exceptional accuracy rate
peaking at 95.12%. It is closely pursued by the Decision Tree (DT) classifier, which demonstrated a
commendable accuracy of 90.24%. These classifiers distinguish themselves by their adeptness in
discerning the most pivotal features within the dataset, those intricately linked to each distinct class

. Conclusively, the model undergoes a comprehensive evaluation through a robust 10-fold cross-
validation (CV) approach. This assessment methodology relies on resampling, which involves
partitioning the data into distinct folds for both training and testing purposes across various iterations.
This rigorous process ensures that the model's accuracy remains reliable and pertinent when applied to
previously unseen samples.

Declarations
Acknowledgment

The authors extend their appreciation to the Deanship of Scientific Research at King Khalid University for
funding this work through a large group research project under grant number RGP2/246/44.

Data availability statement:

The datasets used and/or analyzed during the current study are available from the corresponding author
on reasonable request.

References
1. U. R. Acharya, J. E. W. Koh, Y. Hagiwara, J. H. Tan, A. Gertych, A. Vijayananthan, N. A. Yaakup, B. J. J.
Abdullah, M. K. B. M. Fabell, and C. H. Yeong, “Automated diagnosis of focal liver lesions using
bidirectional empirical mode decomposition features,” Computers in biology and medicine, vol. 94,
pp. 11-18, 2018.
2. N. Chalasani, Z. Younossi, J. E. Lavine, M. Charlton, K. Cusi, M. Rinella, S. A. Harrison, E. M. Brunt,
and A. J. Sanyal, “The diagnosis and management of nonalcoholic fatty liver disease: practice
guidance from the American Association for the Study of Liver Diseases,” Hepatology, vol. 67, no. 1,
pp. 328-357, 2018.
3. J. HAN, M. KAMBER, and J. PEI, "Data Mining: Concepts and Techniques, the Third Edition University
of Illinois at Urbana Champaign," ed: Published by Elsevier, 2012
4. Wang, S., Li, D., Song, X., Wei, Y., & Li, H. (2011). A feature selection method based on improved
fisher’s discriminant ratio for text sentiment classification. Expert Systems with Applications, 38(7),
8696-8702.
5. H M. Abdelaal, and H. Youness, " Knowledge Discovery in the Hadith According to the Reliability and
Memory of the Reporters Using Machine Learning Techniques, " IEEE Access, Digital Object Identifier
10.1109/ACCESS.2019.2944118, 2019

Page 12/20
 6. Gaber, A., Youness, H. A., Hamdy, A., Abdelaal, H. M., & Hassan, A. M. (2022). Automatic Classification
of Fatty Liver Disease Based on Supervised Learning and Genetic Algorithm. Applied Sciences, 12(1),
521, doi.org/10.3390/app12010521.
7. Manjunath, R.V., Ghanshala, A. & Kwadiki, K. Deep learning algorithm performance evaluation in
detection and classification of liver disease using CT images. Multimed Tools Appl 83, 2773–2790
(2024). https://doi.org/10.1007/s11042-023-15627-z
8. Elias Dritsas and Elias Dritsas, “Supervised Machine Learning Models for Liver Disease Risk
Prediction”. Machine and Deep Learning in the Health Domain 12(1), 19; (2023) .
https://doi.org/10.3390/computers12010019.
9. Khan, M.A.R., Afrin, F., Prity, F.S. et al. An effective approach for early liver disease prediction and
sensitivity analysis. Iran J Comput Sci 6, 277–295 (2023). https://doi.org/10.1007/s42044-023-
00138-9
10. Singh, G., Agarwal, C., Gupta, S. (2022). Detection of Liver Disease Using Machine Learning
Techniques: A Systematic Survey. In: Balas, V.E., Sinha, G.R., Agarwal, B., Sharma, T.K., Dadheech, P.,
Mahrishi, M. (eds) Emerging Technologies in Computer Engineering: Cognitive Computing and
Intelligent IoT. ICETCE 2022. Communications in Computer and Information Science, vol 1591.
Springer, Cham. https://doi.org/10.1007/978-3-031-07012-9_4
11. https://mro.massey.ac.nz/server/api/core/bitstreams/d70bb49f-bd1f-455d-bac3
5c128194b338/content (Last Accessed: 17-04-2024)
12. M. Singh, S. Singh, and S. Gupta, “An information fusion based method for liver classification using
texture analysis of ultrasound images,” Information Fusion, vol. 19, pp. 91-96, 2014.
13. S. Bassinet, A. Madani, M. Al-Sarem, and M. Kissi, ‘‘Feature selection using an improved Chi-square
for Arabic text classification,’’ J. King Saud Univ., Comput. Inf. Sci., to be published.
14. J. Han, M. Kamber, and J. Pei, Data Mining: Concepts and Techniques, 3rd ed. Urbana, IL, USA: Univ.
Illinois Urbana Champaign, 2012.
15. Jiawei Han, Micheline Kamber, and Jian Pei, "Data Mining: Concepts and Techniques", the Third
Edition University of Illinois at Urbana Champaign, 2012 Elsevier Inc.
16. HM. Abdelaal, and H. Youness, "Classification of hadith according to its content based on supervised
learning algorithms," IEEE Access, vol. 7, pp. 152379-152387, 2019.
17. https://machinelearningmastery.com/naive-bayes-for-machine-learning/ (Last Accessed: 18-04-
2024) , Jason Brownlee on August 15, 2020 in Machine Learning Algorithms.
18. https://www.geeksforgeeks.org/multinomial-naive-bayes/ (Last Accessed: 18-04-2024)
19. https://www.baeldung.com/cs/decision-tree-vs-naive-bayes (Last Accessed: 18-04-2024)
20. Hastie, T., Tibshirani, R., & Friedman, J. (August 2009). The Elements of Statistical Learning: Data
Mining, Inference, and Prediction. 2nd Edition, Springer New York, NY, ISBN: 978-0-387-84857-0, pp.
XXII- 745 , https://doi.org/10.1007/978-0-387-84858-7

Page 13/20
Figures

Figure 1

See image above for figure legend

Figure 2

See image above for figure legend

Page 14/20
Figure 3

See image above for figure legend

Page 15/20
Figure 4

See image above for figure legend

Page 16/20
Figure 5

See image above for figure legend

Figure 6

Page 17/20
Outline of Proposed System for Liver Disease Prediction

Figure 7

See image above for figure legend

Figure 8

Page 18/20
See image above for figure legend

Figure 9

See image above for figure legend

Figure 10

Analysis of Classifier’s Performance Accuracies (Training Phase)

Page 19/20
Figure 11

See image above for figure legend

Page 20/20