Exploração e comparação de algoritmos de classificação em Machine Learning: uma abordagem estatística

Abstract

This work studies the relationship between Statistics and Machine Learning, specifically in the context of Classifiers Methods, where the computer must learn statistical and computational patterns from the provided data and be able to classify new data based on its learning. The database analyzed in this study contains information about patients with or without heart diseases, and the goal of the methods is to classify new patients as either having or not having the disease. The classifiers chosen for this work were Naive Bayes, K-Nearest Neighbors, and Random Forest. Performance was measured using statistical metrics such as accuracy, specificity, and sensitivity. Additionally, the execution time of each classifier was also measured. In the end, it was observed that Random Forest achieved the best accuracy and specificity, despite other classifiers showing similar results, but it had the worst execution time result. It can be concluded that the selection of the best model may be subjective, as it should take into consideration the application context and the available computational power.