Performance Analysis of Bayesian Learning Against Other Conventional Machine Learning Techniques

Abstract

This study comprises of performance analysis of variants of Naive Bayes and other conventional machine learning algorithms. The mathematics and core intuition behind the machine learning algorithms has been discussed in this text. Models have been trained and tested on different datasets having different genre of features such as continuous, categorical and discrete. Performance analysis has been done to see which model performs better on which dataset. Feature engineering techniques have also been employed and analysis has been done to infer how they impact model’s performance. Breast cancer dataset, heart attack dataset and air quality dataset have been used for the purpose of analysis of results of models on data having continuous independent features. Customized versions of the Naive Bayes algorithms have also been developed by taking the likelihood of features from the probability density functions of Lognorm, Student’s t and Skewnorm distributions. For analysis on data having categorical input features, datasets of Tic-Tac-Toe game, Car evaluation and Connect-4 Game have been used. As far as the discrete features are concerned, Fashion MNIST dataset having pixel values of images as features has been used. An email spam classifie