Analytical Analysis of Five Machine Learning Implementations for Patient Treatments Classification

Abstract

The study reports a wide-ranging comparative review of five machine learning (ML) implementations applied to one patient treatment dataset with the same classification task: determining treatment type from patient information. Methodological differences, model design, preprocessing techniques, and performance results are reviewed to determine best practice and real-world insight for practitioners and researchers. Through controlled benchmarking, we contrast traditional models (e.g., decision trees, logistic regression) with ensemble and neural network ones, examining trade-offs in accuracy, complexity, interpretability, and computational complexity. Our results demonstrate that stacking classifiers and neural networks tend to perform better than simpler models at an accuracy of 73–75%, though in some cases sacrificing explainability and training time. The research also recognizes some of the common issues including class imbalance, feature selection methods, and constraints in cross-validation and hyperparameter tuning. From these observations, we suggest practical recommendations for model choice, dataset preprocessing, and future studies. Our contribution lies in synthesizing practical and methodological insights from five parallel implementations, offering guidance to ML practitioners working on structured healthcare data and extending discussion to generalizable patterns relevant to similar domains