Enhancing Predictive Accuracy in Machine Learning: Techniques for Model Optimization and Feature Selection"s

Abstract

Listen

Translate article

Artificial intelligence (AI) is very relevant in areas like healthcare, finance and e-commerce where a close estimate is crucial in decision making. There are still problems with performance even with the recent developments of Machine Learning models, which stem from insufficient preprocessing, unsuitable feature selection, and poor hyperparameter optimization, which restrict their applicability to multiple domains. This research aims at developing a systematic, step-by-step, and stage-wise improvement of the accuracy of the ML models through data preprocessing and feature selection, and model selection. This study evaluates techniques such as normalization, Recursive Feature Elimination (RFE), and Bayesian hyperparameter tuning by applying this approach to datasets in the healthcare, finance, and e-commerce domains. The findings show that preprocessing increases the accuracy by 5-8%, while RFE maintains 95% of the accuracy with a feature reduction of 30-50%, Bayesian optimization also increases the accuracy by 10-15%, making the overall accuracy of the models to be 96%. This work underscores the importance of the proposed integrated approach for constructing reliable, explainable, and scalable ML models for various fields. The results of the study provide a clear and easily replicable approach useful for future studies in the field and for industries that require high levels of accuracy and model parsimony.