DECISIVE ANALYSIS OF MULTIPLE LOGISTIC REGRESSION APROPOS OF HYPER-PARAMETERS

Abstract

Machine learning based predictive models are playing a vital role in every domain; however, to get the best performance of any model, hyperparameters of the underlying algorithm need to be fine-tuned. Tuning these hyperparameters is an extensive task in terms of resources and time. In this study, we tried to generalize the hyperparameters of the Multiple Logistic Regression model using the iterative method applied on multiple datasets to obtain the best hyperparameters values. In most cases, Newton-cg was found as the best solver with ridge regression. LBFGS and Liblinear were found suitable for a few datasets with L2 regularization irrespective of the linear separability of the data. This work gives the behavioral analysis of hyperparameters of the underlying dataset with respect to Multiple Logistic Regression Classification. It will help the researchers to come up with a more robust generalized tool for fine-tuning of hyperparameters to save the resources.