A Lightweight and Interpretable DepGraph-based Pruned Artificial Neural Network for Advertisement Click Prediction

The rapid expansion of online advertising has necessitated the development of accurate and efficient models for advertisement click prediction (ACP). Predicting whether a user will click on an ad is crucial for optimizing digital marketing strategies, enhancing user engagement, and maximizing revenue. Traditional machine learning approaches, such as KNN, DT, and GBDT, have been widely used for ACP, but they often require extensive feature engineering and lack adaptability to dynamic user behaviors. While deep learning models, particularly artificial neural networks (ANNs), offer improved accuracy by learning feature representations automatically, they suffer from high computational costs and poor interpretability, limiting their practical deployment. To address these challenges, this study proposes a lightweight and interpretable ACP framework that integrates DepGraph-based pruning with SHAP analysis. By leveraging structured pruning techniques, we significantly reduce the model’s computational overhead while maintaining high predictive performance. The use of SHAP ensures interpretability by identifying key features influencing ad clicks, providing actionable insights for advertisers. Experimental results demonstrate that the pruned ANN model retains strong classification capabilities while achieving a substantial reduction in model size and inference time. Moreover, SHAP analysis highlights the most influential features, such as daily internet usage and time spent on site, which play a crucial role in predicting user engagement. This research presents a scalable and interpretable solution for ACP, making deep learning models more efficient for real-world applications in digital advertising.