Cultivar Prediction of Target Consumer Class Using Feature Selection with Machine Learning Classification

Abstract

Recently, Industries are focusing on cultivar prediction of customer classes for the promotion of their product for increasing the profit. The prediction of customer class is a time consuming process and may not be accurate while performing manually. By considering these aspects, this paper proposes the usage of machine learning algorithms for predicting the customer cultivar of Wine Access. This paper uses multivariate Wine data set extracted from UCI machine learning repository and is subjected to the feature selection methods like Random Forest, Forward feature selection and Backward elimination. The optimized dimensionality reduced dataset from each of the above methods are processed with various classifiers like Logistic Regressor, K-Nearest Neighbor (KNN), Random Forest, Support Vector Machine (SVM), Naive Bayes, Decision Tree and Kernel SVM. We have achieved the accurate cultivar prediction in two ways. Firstly, the dimensionality reduction is done using three feature selection methods which results in the existence of reasonable components to predict the dependent variable cultivar. Secondly, the prediction of customer class is done for various classifiers to compare the accuracy. The performance analysis is done by implementing python scripts in Anaconda Spyder Navigator. The better cultivar prediction is done by examining the metrics like Precision, Recall, FScore and Accuracy. Experimental Result shows that maximum accuracy of 97.2% is obtained for Random Projection with SVM, Decision Tree and Random Forest Classifier.