COMPARATIVE EVALUATION OF MACHINE LEARNING ALGORITHMS FOR DISEASE PREDICTION

Abstract

Machine learning has become a key tool in healthcare domains, especially for early detection of pathologies and aiding in the diagnostic process. This research evaluates the comparative efficiency of 6 predictive models, specifically Naive Bayes(NB), Random Forest(RF),Decision Tree(DT),Support Vector Machine(SVM),Logistic Regression(LR), KNearest Neighbour(KNN). Five medical datasets were used to evaluate the models: Heart Disease, Chronic Kidney Disease(CKD),Breast Cancer Wisconsin Diagnostic(BCWD),Hepatitis C(HCV) and Cirrhosis Patient Survival(CPS) Prediction. Feature selection techniques were used to enhance the prediction performance; these techniques are filter methods for feature selection like Chi-square, Mutual Information, NOVA F-test and Correlation analysis. Different train-test split ratios 60/40,70/30 and 80/20 and cross validation techniques (5-fold,7-fold,10-fold) were used for the evaluation of the model performance using precision, accuracy,F1-score and Menthes resulted in a better performance for Random Forest on most of the datasets and best feature selection methods was Mutual Information. Model performance evaluation shows that an accuracy of approximately (63.40%),precision of (63.40%) and f1-score of about(63.14%) were achieved, while the RMSE was around 0.96.The overall ROC-type performance remained balanced and stable across different validation splits,as observed in the evaluation results provided. The results of the classification were near perfect for Chronic Kidney Disease, but were most difficult for Heart Disease. The result underline significance of feature selection and comparably evaluating for making effective and reliable prediction models in healthcare.