A comprehensive study on the performance of different Multi-class Classification Algorithms and Hyperparameter Tuning Techniques using Optuna

Abstract

Multi-class classification talks about classification tasks that have three or more classes. It takes the assumption that every data sample in the dataset is assigned to one and only one class. It is a key idea in machine learning that is used in a variety of applications. Machine Learning includes different types of multi-class classification algorithms like SVM, Decision Tree, Random Forest, etc. This paper will discuss the workings of these algorithms and also emphasize the performance analysis of these algorithms. For performance analysis, this paper takes four multi-class datasets, namely the fetal-health classification dataset, the LED display dataset, the mobile price classification dataset, and the E-Coli dataset. For analyzing the performance of above mentioned three algorithms, this paper uses a machine-learning automated tool for hyperparameter optimization called OPTUNA. Different hyperparameter tuning techniques like Tree-structured Parzen Estimator (TPE) optimization, Covariance Matrix Adaptation Evolution Strategy (CMA-ES) optimization, Random Search, Grid Search, etc. were used for analysis purposes.