Data Mining Techniques for Consumer Behavior Analysis in Digital Fashion Markets

TLDR

The experimental outcomes indicate that the CNN-LSTM model is better than other models, with an accuracy of 98.56, a precision of 98.25, a recall of 99.87, and the F1score of 99.56, which suggests that it is highly effective in impressing the complicated patterns of consumer behavior.

Abstract

The success of any organization is contingent upon the behavior and incidents of its customers. Moreover, consumer purchasing patterns are essential in ensuring the fast expansion of e-commerce sites, as they will aid in retention and targeted advertising. The study examines behavioral patterns and predicts potential churners using the E-commerce Customer Churn Prediction dataset. Some models, such as the Logistic Regression (LR) and Generative Adversarial Network (GAN), Deep Neural Network (DNN), and the proposed CNN-LSTM model, are applied and tested with the help of some of the main performance indicators, such as accuracy, precision, recall, and F1-score. The dataset is preprocessed through missing value are fixed, one-hot encoding is done, the data are normalized, and the balancing of the classes is performed with the help of SMOTE in order to increase the reliability of the models. The experimental outcomes indicate that the CNN-LSTM model is better than other models, with an accuracy of 98.56, a precision of 98.25, a recall of 99.87, and the F1score of 99.56, which suggests that it is highly effective in impressing the complicated patterns of consumer behavior. The outcomes reveal the usefulness of DL methods in projecting customer churn and aiding data-driven decision-making in e-commerce websites of digital fashion.