Ad click fraud has emerged as a major challenge in digital advertising ecosystems, leading to significant financial losses for advertisers and affecting the reliability of pay-per- click (PPC) models. This paper proposes a hybrid machine learning approach for effective ad click fraud detection by com- bining Random Forest and Long Short-Term Memory (LSTM) networks. The proposed system incorporates advanced feature engineering techniques, including click count per IP, time gap analysis, and temporal feature extraction, to improve detection performance. A weighted voting mechanism, defined as 0.7 × Random Forest + 0.3 × LSTM, is employed to integrate the predictions of both models. Furthermore, SHAP (SHapley Additive exPlanations) is integrated to provide explainable AI capabilities, enabling transparent interpretation of the factors influencing fraud predictions. Experimental evaluation on a real- world dataset demonstrates that the hybrid model achieves an accuracy of 97.6% with a fraud recall of 71%, outperforming the individual models. The explainability analysis reveals that click count ip, time gap, and app are the most influential features in identifying fraudulent clicks. The results confirm that the proposed hybrid model offers both high detection performance and interpretability, making it suitable for real-world ad fraud prevention systems.
M.Niharika et al. (Thu,) studied this question.