Filter Based Feature Selection Anticipation of Automobile Price Prediction in Azure Machine Learning

To avoid the problem of overfitting, feature selection is one of key phase in the machine learning process. Feature selection is a quick and easy approach to choose most valuable characteristics for predicting or classifying a target variable in a dataset. They help minimize dimensionality, saving memory and computing time. There are seven filter-based feature selection techniques accessible in Azure machine learning studio (classic) (Spearman Correlation, Fisher score, Pearson Correlation, Count based, Kendall Correlation, Chi squared, Mutual Information). An experiment was developed using Bayesian linear regression and used on the Automobile Price Prediction dataset. The elbow approach was used to pick features based on scores derived from each feature selection technique. When evaluation metrics for each approach were evaluated, it was discovered that Fisher Score was the most efficient.