Computational Modeling of Neural Networks for Predicting Consumer Behavior in Digital Marketing, Application in human resource management for training sales staff

Abstract

This study aims to design and evaluate a computational model based on multilayer perceptron (MLP) neural networks for predicting consumer purchasing behavior in digital marketing. Data from 50,000 users on a digital marketing platform were collected, including demographic variables (age, gender, location, device type), behavioral variables (click count, time spent, return rate, purchase history), and interactional variables (engagement with ads and content). The MLP model was designed with three hidden layers (64, 32, and 16 neurons) and ReLU activation function, and trained using the Adam optimization algorithm with a learning rate of 0.001. The dataset was split into 70% training, 15% validation, and 15% testing, with Dropout and batch normalization applied to mitigate overfitting. For interpretability, SHAP analysis and feature importance assessment were employed. The findings indicated that the proposed model achieved excellent predictive performance (Accuracy = 0.931, F1 = 0.922, Precision = 0.918, Recall = 0.926, ROC-AUC = 0.974). The performance difference between training and test sets was less than 1.5%, demonstrating high model generalizability. SHAP analysis revealed that three features— “purchase history” (24.6%), “number of ad clicks” (18.9%), and “time spent on site” (15.7%)—contributed most significantly to the prediction of purchase likelihood. Moreover, applying the model in simulated scenarios resulted in an average increase in advertising return on investment (ROI) of up to 15% and a reduction of 12.3% in costs due to inappropriate targeting. These results suggest that the designed neural network model can accurately identify complex purchasing behavior patterns and provide valuable managerial insights for data-driven decision-making in digital marketing.