Comparative Analysis of Feature Selection Techniques for Blood Sample-Based Multiple Disease Diagnosis Using Machine Learning

Omitting feature selection leads to large increases in prediction error for disease diagnosis models from machine learning. The proposed study implies the comparison of two most widely used feature selection methods, i.e., SelectKBest (SKB) and Recursive Feature Elimination (RFE) on machine learning models using a blood sample dataset. It selects a small set of 15 features out of 24 are first and later tests the models and bases the evaluation for the models on accuracy. The analysis after the results demonstrates that feature selection has a strong effect on the predictive performance and offer suggestions on how diagnostic accuracy can be improved. Accuracy is the main focus while evaluating the models using various metrics for performance. So, results show that feature selection does make a huge difference in classification performance, and indeed, the SelectKBest does outperform RFE in terms of predictive accuracy, though by a small margin. The results show that ensemble models, particularly the Voting Classifier, showed better performance in disease detection when multiple classifiers are combined.