Predicting E-commerce Customer Purchase Behavior Using LSTMAttention Neural Networks and Data Optimization Strategies

Abstract

In this study, based on neural network algorithm, a prediction model of e-commerce customer purchase behavior was constructed, and the data pre-processing, training strategy and model selection were optimized to improve the prediction accuracy and generalization ability. Firstly, data preprocessing methods such as missing value filling, normalization, Thermal coding and PCA dimensionality reduction are used to effectively optimize data quality and improve the learning ability of the model. Secondly, in terms of training strategy, dynamic learning rate adjustment, gradient clipping, batch training optimization and stop in advance are introduced to reduce overfitting risk and improve training stability. The study compared the performance of logistic regression, random forest, LSTM and LSTM+Attention models. LSTM+Attention model was evaluated using a 12-month e-commerce dataset comprising 1,058 user records, incorporating features such as user demographics, historical purchase behavior, browsing data, shopping cart actions, product metadata, and promotional interactions.The input features were preprocessed using normalization, one-hot encoding, and PCA, and models were trained with optimized hyperparameters using stratified sampling and cross-validation. The study compared four models— logistic regression, random forest, LSTM, and LSTM+Attention.LSTM+Attention achieved the highest performance with 91.3% accuracy and AUC-ROC of 0.95, outperforming logistic regression (76.8%), random forest (81.5%), and LSTM alone (88.9%).These results demonstrate the effectiveness of combining temporal modeling with attention mechanisms in predicting e-commerce customer purchase behavior. This paper analyzes the value of predictive model in business applications such as personalized recommendation, precision marketing, dynamic pricing, inventory management, etc., and shows that the model can effectively improve the operational efficiency and user experience of e-commerce platforms. Through literature review, this study further discusses the development trend of e-commerce prediction model, which provides theoretical support and practical guidance for the intelligent decision-making of e-commerce platform.