Feature Selection Techniques to Predict the Religion of a Country from Its Flag

Abstract

Feature selection is a process of preparing data to be more effective and efficient for machine learning problems. The purpose of feature selection is to select relevant features from huge number of features. To build a simple model which will be easy to understand data and take less time to train the model, thereby optimizing model performance. The paper proposes two feature selection techniques namely Lasso and Select From Model (meta-transformer) to select relevant features from flag dataset that intensifies the model performance. For the prediction of religion of a country, three tree-based classifiers are used—random forest, decision tree, and extra trees model. Among these, random forest classifier gives best prediction.