Design of schemes For feature selection using sensitivity and correlation analysis in supervised Classification

Abstract

A feature is a distinct, measurably relevant aspect of the process newlineunder observation. Using a set of features, any machine learning newlinealgorithm can be trained for classification. In recent years, the number of newlinefeatures used for model training have grown exponentially and some of newlinethe model trainings have been trained using thousands of features. Several newlinefeature selection strategies have been developed to solve the challenge of newlineremoving irrelevant and redundant features that slows down the training newlineand impacts the performance of the model. Work on identifying optimum newlinefeatures makes a major contribution in improving classifier performance, newlinereducing computation cost and better understanding of the data in newlinemachine learning applications. newlineMost of the research about the feature selection predominantly newlinefocusses on reducing the feature but none of them can be considered as newlinebest method to identify the significant features. As a result, new feature newlineselection methods are continuously emerging by integrating various newlinetechniques of feature selection which combines more than one feature newlineselection approaches, developing new ways for improving classifier newlineperformance. newlineFor building a classification model, feature selection is an newlineessential step. The best features can lead to the most accurate newlineclassification results. Also, features count can directly impact the newlinecomputational time and model performance of the classification process newlineand hence it should be optimized. newlinevi newlineThe traditional way of selecting best features are wrapper, filter newlineand embedded method. All three approaches were performed using either newlinea single or a combination of two approaches. Several existing works have newlinebeen proposed to find the significant features to give better model newlineperformance. But in most of the works, the best set of features are newlineidentified only after establishing a machine learning model. Identifying newlinethe optimum features before passing all the features to the ML model is a newlinemajor challenge however it will help in model maintenance and newlinesubsequent retraining. newline