PROJECT STAGE-1 REVIEW PAPER

LIVER DISEASE DETECTION USING ML

K.MANIDHEEPA 21261A6723
PRACHI JAIN 21261A6746

Abstract

Liver disease is a significant public health concern worldwide, with millions of cases diagnosed every
year. Early detection of liver disease is crucial, but traditional methods for diagnosis can be slow and
prone to error. The use of machine learning algorithms for predicting liver disease has become
increasingly popular, providing more accurate and efficient results. This review paper explores the
implementation of two widely-used algorithms—Support Vector Machines (SVM) and Naive Bayes—for
the prediction of liver disease. The analysis is based on a liver function dataset with attributes like enzyme
levels and protein counts. The comparison of the two algorithms is presented in terms of accuracy,
precision, recall, and computational efficiency. Results show that while both algorithms have their merits,
SVM consistently outperforms Naive Bayes in prediction accuracy. However, Naive Bayes offers
simplicity and speed, making it a viable option in resource-constrained environments. Future work could
focus on optimizing these models further to ensure their applicability in real-world clinical settings.

Keywords Liver disease prediction, Support Vector Machines (SVM), Naive Bayes, Machine learning,
Liver function tests, Classification, Healthcare diagnostics.

Introduction

The liver is a vast, substantial organ in the human body. Liver damage is the one of the top deadliest disease in
Weighing around 3 pounds. The liver contains two vast the world. The main causes of liver damage are Fatty
segments, called the right and the left projections. The liver, Liver Fibrosis, Cirrhosis, hepatitis and infections
gallbladder sits under the liver, alongside parts of the [2]. Fig 1. Shows the stages of liver damage, in the first
pancreas and digestive organs. The liver and these stage healthy liver will become fatty liver due to
organs work together to process, ingest, and process accumulation of cholesterol and triglycerides, after few
suste nance. The liver's main role is to filter the harmful months to years fatty liver will becomes liver fibrosis,
substances in the blood originating from the digestive later it leads to final stage of liver damage known as
system, before passing it to the rest of the body [1]. cirrhosis.
In the early stages of the liver disease, it is very difficult to identi fy even though liver tissue has been damage
moderately, it origins many medical experts repeatedly fail to diagnose the disease. This can distort to wrong medicine
and treatment, so early detection is very important and necessary to save the patient.

The liver is an essential body organ that forms an induced by the intrusion of the HBV genome[5]. The
important barrier between the gastrointestinal blood, important thing in preventing liver cancer is to prevent
which contains large amounts of toxins and antigens in hepatitis virus infection and eliminate hepatitis virus in
the body. The liver produces a large amount of chronic hepatitis patients. Automatic classification tools
hormones, enzymes, and performs several functions may reduce the burden on doctors. Data classification is
essential to life. It is also the organ responsible for a two phase process in which first step is the training
cleansing of toxins from the bloodstream, by turning phase where the classifier algorithm builds a classifier
them into removable substances. Liver diseaserefers to with the training set of dataset the second phase is
many diseases and disorders that may cause impaired classification phase where the model is used for
liver function that can make liver decrease of its classification and its performance is analyzed with the
functions. The dysfunction may be primary, but the liver testing set of datasets[6]. Existing feature selection
is often secondarily affected by disorders of other organ methods broadly fall into two categories, filter methods
systems, since it is involved in many metabolic and and wrapper methods. Filter methods select features
detoxifying processes. Hepatic fibrosis and its end stage based on some discriminate criteria that rely on the
cirrhosis are an increasing worldwide concern. Cirrhosis characteristics of data and are independent of any
is the irreversible end result of fibrosis scarring and classification algorithms[7]. Wrapper methods use the
normal hepatic architecture is replaced by predictive accuracy of predetermined classification
interconnecting bands of fibrosis tissue. The most algorithms as the criteria to determine the goodness of a
common etiological factors resulting in cirrhosis are subset of features[8, 9]. Most wrapper methods adopt
hepatitis B, hepatitis C, and excessive alcohol sophisticated multivariate machine learning tools such
consumption disease that may produce few or no as SVMs that take the combinatorial effects of features
symptoms for many years after infection. Some patients into account. These have been shown in many
develop chronic infection and suffer no significant liver experiments to be more powerful in terms of
damage, while others progress quickly to liver cirrhosis classification accuracy than the filter methods[10].
and may develop hepatocellular carcinoma[2]. Patients Support Vector Machines proved to be effective for a
with chronic liver diseases belong to a high-risk group lot of classifications problems. For binary-class
for hepatocellular carcinoma and should be followed up classification, SVM constructs an optimal separating
regularly for early diagnosis. Chronic HCV infection is hyper plane between the positive and negative classes
the major cause of cirrhosis and hepatocellular with the maximal margin. It can be formulated as a
carcinoma (HCC). In this condition, alpha fetoprotein quadratic programming problem involving inequality
levels may be elevated. The incidence of hepatocellular constraints[11, 12]. SVMs are one of the most
carcinoma is rising, and this trend is expected to promising machine learning algorithms and there are
continue for years[3]. Figure 1, show that liver cancer is many examples, where SVMs are used successfully, e.g.
the most cause of death in Egypt among other types of text classification, face recognition, and Bioinformatics.
cancer. According to the current studies, the majority of On these data sets SVMs perform very well and often
HCC patients contracted the disease from the outperform other traditional techniques[13]. SVMs have
accumulation of genetic abnormalities, probably gained an enormous popularity in statistics, learning
induced by exterior etiological factors especially HBV theory, and engineering[14, 15], and the many
and HCV infections[4]. These risk factors can induce references therein. With a few exceptions most support
mutations and damage in DNA sequences, such as p53 vector learning algorithms have been designed for
mutation induced by aflatoxin and DNA damage binary
 pyruvic transaminase (SGPT), andalkaline phosphatase
problems. A few attempts have been made to generalize (Alkphos) using two datasets. The first one is BUPA
SVM to multiclass problems[16]. In this work, a support Liver Disorders datasets taken from the University of
vector machine is used as a classification algorithm in California at Irvine (UCI) Machine Learning
order to compare its performance with different features Repository, and the second one is from ILPD (Indian
combinations such as aspartate aminotransferase Liver Patient Dataset), it was collected from north east
(SGOT), glutamic of Andhra Pradesh, India

MATERIALS AND METHODS

The liver is one of the major targets for insulin and its count regulatory hormones, such as glucagon. HCC
patients who abuse alcohol are more likely to develop cirrhosis than those who do not. The most common
cause of liver disease is non-alcoholic fatty liver disease. Cirrhosis is the end-result of many liver conditions
and involves severe scarring of the liver. It is associated with a progressive decline in liver function resulting
in liver failure. Hepatocellular carcinoma is the most common primary cancer of the liver. There are factors
that may impact progression include age, gender, chronic alcohol abuse, and quantity of virus of exposure.
The disease appears to be more aggressive in patients that acquire HCV after age 40 and may be more
progressive in men than women[2]. In this paper SVM classification algorithm has been applied to:
BUPA liver disorders dataset and Indian Liver
Patient Dataset for evaluating SVM performance with different features.
 Research by [Author et al., Year] showed
MACHINE LEARNING ALGORITHMS FOR WSVM to significantly improve
LIVER DISEASE DETECTION classification accuracy for liver disease
datasets with high class imbalance.
(1) SUPPORT VECTOR MACHINES (SVM)  Fuzzy SVM (FSVM) : FSVM integrates a
Support Vector Machines (SVM) are particularly useful fuzzy membership function to reduce the
for classification tasks with high-dimensional data, making impact of noisy or uncertain data points,
them ideal for medical diagnostics. SVM constructs a which are common in medical records.
hyperplane that best separates data points into different Studies [Author et al., Year] have shown
categories. In liver disease prediction, the input features— FSVM to yield robust performance in liver
such as enzyme levels and other liver function metrics— disease detection, where minor variations
are often non-linearly separable. SVM, through kernel in features are common.
functions like Radial Basis Function (RBF) and  Hybrid Approaches: GA-SVM and
Polynomial kernels, effectively handles such complexities. PSO-GA-SVM: Genetic Algorithm (GA)
and hybrid PSO-GA-SVM approaches
TYPES OF SVM: optimize the hyperparameters of SVM by
 Standard SVM for Liver Disease combining evolutionary strategies.
Detection: SVM is widely applied due to Literature indicates that these methods
its capability to handle complex achieve high accuracy due to their
relationships between features. Studies exploration capabilities and resistance to
such as [Author et al., Year] have local minima, which can trap traditional
demonstrated that RBF-kernel SVM optimization methods [Author et al.,
yields high classification accuracy with Year].
relatively minimal data preprocessing. Application to Liver Disease Detection:
Parameter Optimization: The choice of C Hybrid methods are computationally
and gamma parameters in RBF-SVM has intensive but have shown promise in
a significant impact on performance. achieving high accuracy and minimizing
Techniques like Grid Search and Cross- classification error, making them suitable
Validation (CV) are common in studies to for medical diagnostics.
optimize these parameters for specific
datasets. ADVANTAGES OF SVM:
 Particle Swarm Optimization (PSO)-  High Accuracy: SVM is particularly robust in
SVM: high-dimensional spaces, where the relationship
PSO is an evolutionary algorithm that between features may not be linear.
can optimize SVM parameters by  Kernel Flexibility: By utilizing different
iteratively improving candidate solutions. kernels, SVM can adapt to the underlying structure
Research by [Author et al., Year] found of the data, which enhances its ability to predict
PSO-SVM to outperform traditional SVM disease outcomes.
methods in detecting liver disease,  Regularization: SVM includes regularization
especially in imbalanced datasets, by techniques to prevent overfitting, which is crucial
efficiently tuning hyperparameters. when working with complex medical datasets.
Advantages: PSO-SVM can converge on CHALLENGES WITH SVM:
an optimal solution more quickly and  Computational cost: SVM can be
may better handle non-linear computationally intensive, especially when applied
relationships in the data, yielding higher to large datasets. This is a critical factor in clinical
accuracy and lower false positive rates settings where real-time results are needed.
than standard SVM.  Model Interpretability: Despite its
 Weighted SVM (WSVM): WSVM is accuracy, SVM’s "black-box" nature makes it
effective for imbalanced datasets, difficult to interpret, which may hinder its
common in medical data where the acceptance by healthcare professionals.
proportion of diseased to healthy
samples is low. By assigning greater
weight to the minority class, WSVM
reduces bias toward the majority class.
 applications, particularly when the training dataset is large.

ADVANTAGES OF NAÏVE BAYES:

 Simplicity: Naive Bayes is straightforward to
implement and requires minimal computational
resources, making it ideal for real-time predictions
in resource-constrained environments.
 Efficiency: Due to its probabilistic nature,
Naive Bayes can quickly process large datasets,
Fig. SVM offering rapid predictions.
Algorithm
CHALLENGES WITH NAÏVE BAYES:
(2) NAÏVE BAYES  Feature Independence Assumption:
The algorithm’s assumption that features are
Naive Bayes is a simple yet effective classifier based on the independent can lead to inaccuracies when the
assumption of feature independence. In liver disease features are correlated, which is often the case in
prediction, Naive Bayes calculates the probability of a medical diagnostics.
patient having the disease based on liver function test  Lower Accuracy: Compared to more
results and other clinical factors. Despite its simplistic sophisticated algorithms like SVM, Naive Bayes
assumption, Naive Bayes often performs well in real-world may underperform in terms of accuracy,
particularly when the dataset is small or noisy.

Fig. Equation of Naïve Bayes algorithm

DATASET

1. BUPA Liver Disorders: BUPA liver disordershas 6 numeric Attributes, 345 Instances. Relevant information:
The first 5 variables are all blood tests which are thought to be sensitive to liver disorders that might arise
fromexcessive alcohol consumption, each line in the BUPA. Data file constitutes the record of a single male
individual. It appears that drinks>5 is some sort of a selector on this dataset. University of California at Irvine (UCI)
machine learning repository (WWW.UCI.Com).
2. Indian Liver Patient Dataset (ILPD): Indian Liver Patient Dataset (ILPD) has 9 attribute, 483
Instances. The data set was collected from north east of Andhra Pradesh, India. Selector is a class
label used to divide into groups (liver patient or not). This data set contains 441 male patient
records and 142 female patient records .this data downloaded from (WWW.UCI.Com).
EVALUATION MEASURES:

RECALL:
Recall can be defined as the proportion of the total number of correctly classified positive examples divide by the total
number of positive examples. High Recall point outs the class is correctly acknowledged (a small number of FN).

PRECISION

To get the estimation of precision we seperate the all out number of effectively characterized positive qualities by
the complete number of anticipated positive qualities. High Precision demonstrates a model delegated positive is
for sure positive (few FP).High review, low precision: This implies that the entirety of the positive models are
effectively recognized (low FN) however there are a great deal of false positives.Low review, high precision: This
shows that we a ton of positive models (high FN) yet those we foresee as positive are really positive (low FP)

ACCURACY
It is an evaluation measure that reflects the overall correctness of a model in liver disease detection. It
calculates the proportion of correctly identified cases both positive and negative—out of the total cases.
The formula for accuracy is:
F-MEASURE: individuals) than positive samples (patients
Since we have two measures (Precision and Recall) it with liver disease). Techniques like Synthetic
assists with keeping up an estimation that speaks to Minority Over-sampling Technique (SMOTE)
them two. We figure a F-measure which is utilized or class-weighted loss functions are
Harmonic Mean instead of Arithmetic Mean as it commonly used to mitigate this issue.
rebuffs the extraordinary qualities more. The F-Measure
will everytime be closer to the littler estimation of CHALLENGES AND RESEARCH
Precision or Recall.The confusion matrix shows the DIRECTIONS:
ways in which your classification modelis confused  Handling Noisy and
when it makes predictions. Incomplete Data: In medical
datasets, incomplete and noisy data are
common. Advanced imputation
techniques like KNN imputation or
Expectation-Maximization (EM) can help
mitigate the effects of missing data, while
noise-tolerant algorithms like SVM with
soft margins can improve model
robustness.
 Real-time Implementation
CLASSIFICATION/RATE ACCURACY: and Scalability: The scalability of
these models in clinical settings is an
ongoing challenge. SVM, despite its
accuracy, is computationally intensive.
Research should focus on developing
more efficient implementations of SVM,
possibly using parallel processing or cloud
computing to enable real-time applications
in healthcare.

DATA PRE-PROCESSING AND

FEATURE SELECTION  Interpretability and
In liver disease prediction, the quality of the input data Transparency: Model interpretability
is crucial. Common preprocessing techniques include is a key concern in medical diagnostics.
normalization, missing value imputation, and outlier While Naive Bayes is more interpretable,
detection. Feature selection is another vital step. SVM’s decision-making process is
Recursive Feature Elimination (RFE) and Principal opaque. Future research could focus on
Component Analysis (PCA) are often used to reduce the enhancing the interpretability of SVM by
dimensionality of the data, focusing on the most integrating it with more transparent
significant features for liver disease prediction. models like decision trees or rule-based
 Impact of Feature Selection on systems.
Model Performance: Selecting the right
features can significantly improve model
accuracy. In liver disease prediction, key
features include albumin levels, bilirubin
counts, and enzyme levels (AST, ALT).
Removing irrelevant or redundant features can
reduce overfitting and enhance the
generalization ability of the model.
 Data Imbalance and Mitigation:
Medical datasets, including liver disease
datasets, often suffer from class imbalance—
there are more negative samples (healthy
LITERATURE SURVEY:

 Elias Dritsas et al. (2023): This study evaluated multiple ML models for liver disease prediction,
emphasizing the effectiveness of a Voting classifier, which outperformed others with an accuracy of
80.1% and an AUC of 88.4%. SMOTE was used to balance data, showing ensemble methods' value in
improving diagnostic accuracy in liver disease prediction

 Lee et al. (2022): Using clinical and radiological data, this research developed an ML model for
predicting liver cancer recurrence, particularly hepatocellular carcinoma (HCC). The model showed
potential to support physicians with accurate predictions, allowing for tailored follow-up treatments

 Ikeda et al. (2022): This study utilized ML to predict survival rates in advanced HCC patients treated
with lenvatinib. The resulting model enabled clinicians to make personalized treatment decisions based
on survival predictions, highlighting ML’s role in personalized medicine.

 Banu and Vasan Durai (2022): This analysis explored K-means and Decision Tree (C4.5) classifiers and
found KNN and SVM to be most accurate for liver disease prediction. Feature selection proved critical
for optimizing model accuracy, and hybrid models were shown to improve prediction robustness

 Pushpendra Kumar et al. (2022): Proposed the Fuzzy-ANWKNN algorithm, which specifically
addresses class imbalance issues in liver disease datasets. Using Liver Function Test (LFT) data, this
study showcased improved predictive accuracy and revealed the importance of data balancing techniques
in ML models for liver disease diagnosis (IEEE XPLORE)

 Wang et al. (2020): Developed an ML model to predict early recurrence of HCC after curative resection.
The model outperformed conventional methods, aiding clinicians in identifying patients at high risk of
recurrence, thus supporting timely intervention and personalized care.

 Lee et al. (2020): Conducted a systematic review on ML algorithms for predicting HCC recurrence post-
liver transplantation. This meta-analysis showed that ML models improve recurrence prediction
accuracy, highlighting potential applications in post-transplant patient monitoring for better outcomes.

 Narwaria et al. (2021): In a systematic review, this study analyzed ML applications in liver cancer
surgery. ML models were shown to support surgical planning and improve outcomes, especially in
decision-making for liver resection and transplantation planning.

 Mafazalyaqeen Hassoon et al. (2019): This study introduced an optimized rule-based approach using the
Boosted C5.0 classifier for diagnosing liver disorders. This model aimed to support timely diagnosis,
helping medical professionals to identify liver disease symptoms faster and more accurately.

 Sanjay Kumar et al. (2019): Explored various ML classification techniques on real-time liver disease
patient datasets. Decision Tree, SVM, and Logistic Regression algorithms were used, with analysis
metrics like precision, recall, and accuracy highlighting the models’ efficiency in predicting liver disease
accurately.

 Pushpendra Kumar et al. (2019): Addressed data challenges like class imbalance in liver disorder
 prediction by employing the Fuzzy-ANWKNN algorithm on LFT datasets. This novel approach yielded
accurate predictions, especially helpful in handling the imbalanced liver disorder data that often skews
conventional classification algorithms.

SIMULATION OF EXISTING SYSTEMS FOR LIVER DISEASE DETECTION AND

THEIR DRAWBACKS.

Machine Learning Models: problems, this causes overfit, which implies models that
work with training data do not work on novel data.
Simulation: Decision Trees, Support Vector Machines Also, training CNNs requires a big amount of
(SVM), and various ensemble methods such as Random computation resources, which turns out to be expensive.
Forests and Voting classifiers have been simulated on
other liver disease datasets as well frequently. Liver Interpretability: CNNs and other deep learning
disease classifiers such as Decision Trees use standard approaches are also termed as black boxes, as
datasets of liver disease with enzyme levels as well as automated predictions do not explain anything. In the
demographic variables, such as the UCI Liver Disorder case of clinicians, this becomes a problem since they
dataset. expect a solid rationale behind a diagnosis or a
prediction before proceeding to trust and execute
the varied outcomes
Drawbacks:

Class Imbalance: Epidemiological surveys concerning

liver disease are usually on case-control studies. Many Feature Engineering and Selection:
of them show a relatively high proportion of healthy Simulation: Researchers often simulate models
individuals in comparison to the diseased ones. Class featuring varying levels of feature engineering in
imbalance leads to biased models perusing the majority an effort to ascertain the most important liver
class at the expense of statistical accuracy in liver disease predicting features such as liver enzymes,
disease detection. Although there are other methods like bilirubin, and even demographic variables.
SMOTE that can add synthetic cases, they are also not Techniques like RFE (Recursive Feature
optimal in effectiveness because of the variability of the Elimination) are quite often adopted to enhance
generated synthetic cases.
model accuracy.
Deep Learning and Neural Networks: Drawbacks:
Simulation: In the case of liver disease classification,
especially when the images in question include medical
Feature Relevance Dependence: A large degree
scans like CT or MRI scans, deep learning methods of dependence on data sources can be seen as a
such as Convolutional Neural networks are commonly significant drawback in terms of generalization.
employed. These models perform well in the For instance, a model trained on Kappa score will
segmentation of the features, which would allow the always return Kappa from that dataset resulting in
reaching to the liver and its pathologies like cirrhosis a highly distorted figure and Kappa rating of that
and fibrosis. particular model. Therefore, features with low
importance will lead to distortion in results.
Drawbacks:
Manual Feature Engineering: As with most ML
Data and Computational Requirements: In order for
models, feature engineering is often considered a
deep learning models to generalize, they need to be
black box. As IoT expands, feature mining
trained using quite amount of data. Because medical
imaging databases are often small because of privacy becomes an extensive and painstaking process,
particularly in the presence of numerous variables
in sizable data sets. that can be shared and studied, thereby resulting in
small, non-diverse datasets that are brittle for ML
GAP ANALYSIS AND DRAWBACKS IN solutions.
LIVER DISEASE DETECTION.
Limitations: Lack of access to data limits the
Class imbalance and limitations of the dataset: generalizability of detection models for liver
diseases. While promising, smart clinical decision
Gap: Datasets for liver disease detection are often support using federated learning — where models
characterized by class imbalance, leading towards learn from decentralized data sources without
over-specialization of the developed ML models on compromising privacy — can be technically
a prevailing class referred to as “healthy” and challenging to implement and resource-demanding
others. Thus, sensitivity in detecting the cases of on the local sites.
liver disease is greatly impaired.
Limitations 4: Real Time and Predictive
Drawbacks: Other methods currently in use such Modelling
as SMOTE have managed to solve some of these
challenges, but not entirely as they leave a Gap: The existing models mainly concentrate on
possibility of models learnt on artificial samples static prediction but hardly dynamic disease
overfitting and delivering poor results on actual progression or treatment response prediction.
data. Prediction models that output predictions in real-
time (like prediction results from continuous data
Interpretability and Explainability in Clinical that is streamed such as daily lab results) are
Settings: uncommon.

Gap: Most of the state of the art algorithms (e.g. Disadvantages: Since it is not real-time and
CNNs, ensemble models) which performed the best predictive, the ML models cannot offer solutions to
have poor explainability making it hard for health management advice proactively. Creating
clinicians to easily understand and rely on these dynamic models that operate in real-time takes a
models particular decisions. lot of computing power and integration with
hospital databases (which is hardly feasible to do at
Drawbacks: Such risks of poor explainability scale).
create a barrier that prevents the clinical
application of these models. There are initiatives Incorporating into Clinical Workflows:
trying to resolve the issues like LIME (Local and
Interpretable Model-agnostic Explanations) and Gap Between Research and Practice: Previous
SHAP (SHapley Additive exPlanations), however, liver disease detection models have been developed
they only increase the complexity of the model in research environments with very limited in vivo
which may not be practical in real-time clinical validation leading to gap between academia-based
scenarios where such decisions would be required development of models and readily adoption into
ayern, D.P. 2013 the clinical practice.

Restricted Data Privacy and Sharing: Disadvantages: The absence of incorporation into
clinical workflows (like EHR systems) makes the
Gap: From GDPR to HIPAA regulations privacy model less applicable for real-time patient service.
has decreased the amount of sensitive patient data Closing this gap will entail more testing and
validation of models between researchers and classification, clustering, association rules
healthcare providers in clinical settings, and so on.
establishing whether they meet clinical needs • The main objective of this research work is
to predict liver diseases using classification
PROPOSED SYSTEM: algorithms.
• The algorithms used in this work are Naïve
• In recent years in healthcare sectors, data Bayes and support vector machine (SVM).
mining became an ease of use for disease • These classifier algorithms are compared
prediction. based on the performance factors i.e.
• Data mining is the process of dredge up classification accuracy and execution time.
information from the massive datasets or From the experimental results it is observed
warehouse or other repositories. that the SVM is a better classifier for
• It is a very challenging task to the predict the liver diseases.
researchers to predict the diseases from the
voluminous medical databases.
• To overcome this issue the researchers use
data mining techniques such as

CONCLUSION:
The comparison between Support Vector Machines and Naive Bayes reveals that each algorithm has its strengths
and limitations. While SVM offers higher accuracy and better handling of complex data relationships, Naive
Bayes provides simplicity and speed. For liver disease prediction, SVM is a preferable choice in settings where
accuracy is paramount, while Naive Bayes could be applied in resource-constrained environments that require
rapid predictions. Future work should focus on enhancing the interpretability of these models and improving their
scalability for real-time clinical applications.

REFERENCES:

1. A. Sharma, S. Gupta, and R. Kumar. “An Integrated Approach to Enhance the

Prediction of Liver Disease.” IEEE Transactions on Biomedical Engineering, vol. 71,
no. 2, pp. 222-230, 2024.

2. S. Kim, H. Hwang, and Y. Lee. “Diagnosis of Liver Diseases using Deep

Learning Algorithms.” Journal of Biomedical Informatics, vol. 98, pp. 103-112, 2024.

3. B. P. Singh, M. Khanna, and S. Grover. “A Comparative Study on Liver

Disease Prediction using Supervised Learning Algorithms with Hyperparameter
Tuning.” IEEE Conference on Machine Learning and Applications (ICMLA), pp. 113-
120, 2023.

4. J. Patel and A. Shah. “Comparative Analysis of Classification Models for Liver

Disease Prediction.” BMC Medical Informatics and Decision Making, vol. 23, pp. 1-9,
2023.

5. R. Torres, M. Lopez, and N. Sinha. “A Machine Learning Framework for

Chronic Liver Disease Diagnosis.” Computer Methods and Programs in Biomedicine,
vol. 231, pp. 106421, 2023.

6. L. Jackson, P. Chen, and K. Davidson. “Predicting Liver Fibrosis Using

Ensemble Machine Learning Approaches.” Journal of Hepatology, vol. 78, no. 5, pp.
1102-1110, 2023.

7. C. Iyer, R. Thomas, and L. Waghmare. “Liver Disease Prediction using Semi-

Supervised Machine Learning Algorithm.” IEEE Transactions on Computational
Biology and Bioinformatics, vol. 20, no. 1, pp. 33-40, 2022.

8. A. Hernandez, M. Garcia, and J. Singh. “Use of Random Forest for the

Diagnosis of Hepatocellular Carcinoma.” Scientific Reports, vol. 13, pp. 21012,
2022.

9. F. Gomes, L. C. Alencar, and J. M. Pereira. “An Ensemble Approach for

Predicting Liver Cirrhosis Stages.” Artificial Intelligence in Medicine, vol. 120, pp.
102168, 2022.

10. K. Bhatt and A. Kumar. “Comparative Study of Deep Learning Algorithms for
Liver Disease Classification.” Computers in Biology and Medicine, vol. 145, pp.
105511, 2022.

11. T. Banerjee, R. P. Singh, and K. Arora. “Early Prediction of Liver Disease Using
Machine Learning Algorithms.” Journal of Translational Medicine, vol. 20, pp. 119,
2022.

12. D. R. Patel, A. Johnson, and H. Zheng. “Liver Disease Prediction Using

Machine Learning Algorithms.” IEEE Access, vol. 9, pp. 148172-148181, 2021.

13. M. Sato and Y. Nakajima. “Artificial Intelligence in Liver Disease Diagnostics:

A Comprehensive Review.” The Lancet Digital Health, vol. 3, no. 9, pp. 643-652,
2021.

14. G. Zhou and Y. Li. “Deep Learning in Predicting Chronic Liver Disease
Progression.” PLoS One, vol. 16, no. 7, pp. e0254160, 2021.

15. J. Clarke, L. Wong, and T. Hsu. “Machine Learning Techniques in Hepatology:

A Systematic Review.” Hepatology International, vol. 15, pp. 597-608, 2021.

16. E. Choudhury and M. Raman. “Application of Deep Learning in Predicting

Liver Diseases: A Case Study.” IEEE Journal of Biomedical and Health Informatics,
vol. 25, no. 10, pp. 2895-2903, 2020.

17. A. T. Nelson, C. M. Roberts, and T. E. Warren. “Neural Network Approaches

for Predicting Liver Function Outcomes.” Journal of Clinical Gastroenterology, vol.
54, no. 6, pp. 346-353, 2020.

18. F. Wei, L. M. Ting, and Q. Zhang. “Optimizing Liver Disease Prediction with
Ensemble Machine Learning Models.” IEEE Transactions on Information Technology
in Biomedicine, vol. 18, no. 4, pp. 612-620, 2019.

19. G. Maramreddy, R. Velpuri, and P. Sandhya. “A Novel Approach for Predicting

Liver Disease Using Neural Networks.” IEEE International Conference on Machine
Learning and Applications, pp. 345-352, 2019.

20. H. Song, K. Li, and J. B. Kim. “Liver Disease Prediction via Feature
Engineering and Classification Techniques.” IEEE International Conference on
Bioinformatics and Biomedicine, pp. 78-85, 2018.
21. U. Choi, M. Hasan, and W. Cho. “A Comprehensive Study on Liver Disease
Detection Using Data Mining.” Computer Methods and Programs in Biomedicine,
vol. 140, pp. 23-33, 2018.

22. R. Singh, T. Roy, and J. Abraham. “Predictive Analysis of Liver Disorders Using
Machine Learning Techniques.” Journal of Medical Systems, vol. 42, no. 8, pp. 157,
2018.

23. N. Gupta, S. S. Anand, and R. Sharma. “Detection of Liver Diseases using

Support Vector Machine and Artificial Neural Network.” IEEE Region 10 Symposium
(TENSYMP), pp. 1-6, 2017.

24. Y. Zhou, L. Wang, and X. Yuan. “Prediction of Liver Disease Risk using
Machine Learning Techniques.” Medical and Biological Engineering and Computing,
vol. 55, pp. 1245-1253, 2017.

25. J. C. Wong, T. M. Lee, and A. S. Martinez. “An Analysis of Liver Disease

Prediction Models Based on Machine Learning.” Bioinformatics, vol. 33, no. 7, pp.
987-993, 2017.

26. M. Ahmed, F. Lim, and J. Wong. “Risk Factor Analysis in Liver Cirrhosis
Patients Using Logistic Regression and Decision Trees.” Journal of Clinical Medicine
Research, vol. 8, no. 2, pp. 110-116, 2016.

27. H. Y. Kim, S. M. Park, and T. K. Hong. “Application of Data Mining in Predicting

Hepatitis and Liver Disorders.” Health Informatics Journal, vol. 22, no. 4, pp. 917-
928, 2016.

28. A. Torres and K. Brown. “A Predictive Model for Liver Disease Detection:
Application of Artificial Intelligence in Medical Diagnostics.” Artificial Intelligence in
Medicine, vol. 68, no. 2, pp. 135-143, 2016.

29. L. Martinez, M. T. Gomez, and F. Sanchez. “Comparative Analysis of Machine

Learning Techniques for Predicting Liver Disease Outcomes.” BioMed Research
International, vol. 2015, Article ID 829132, 2015.

30. S. Chandra, M. Patel, and L. Thomas. “Classification Techniques for Predicting

Liver Diseases Using Electronic Health Records.” Journal of Healthcare Engineering,
vol. 6, no. 3, pp. 127-138, 2015.