A Proposed Framework For Consumer Purchase Behaviours Using Feature Selection Based On SVMs And Random Forest

In marketing and customer analysis, customer characteristics are important indicators that describe customer behaviour, preferences, etc. In addition to offering businesses access to more comprehensive and detailed customer data, a well-performed customer feature filter can enable researchers to perform real-time analysis and make informed decisions. Meanwhile, as the types and sources of data increase, problems such as the curse of dimensionality and noise also arise. Removing noise and reducing dimensions to enhance model performance has become an increasingly important consideration. This research considers feature selection, which involves selecting a set of the most informative features from the original feature space, as a means to significantly reduce noise and reduce dimensions, thereby improving efficiency. Which means the framework can process all types of features and make selections. One such effective feature selection method is RF-RFE (Random Forest Recursive Feature Elimination), which is particularly suited for high-dimensional datasets. RF-RFE combines the power of Random Forest (RF) with Recursive Feature Elimination (RFE), iteratively ranking and eliminating less important features based on the model’s performance, until an optimal subset of features is identified. However, this may lead to a large number of features being selected, or if there is no evidence that these selected features can perform well. It is often ambiguous and subjective regarding final subset selection. This research introduces RFRFE combined with Support vector machine (SVM) models for feature selection. SVM is a machine learning algorithm. In recent years, there has been a vast amount of research on SVM and introduced as a powerful method for classification. The research expects that SVM uses a high dimension space to find a hyperplane boundary to perform binary classification and minimize the error rate. We expect that RF-RFE-SVM may offer competitive or superior performance in terms of precision and recall, particularly in identifying customers who are likely to make a purchase. As research result shows, this approach can significantly improve the model performance compared with using RF-RFE alone.