CNN-LSTM Based Click Fraud Detection System with GUI Interface and Explainable AI

Abstract

Click fraud poses a significant challenge in digital advertising, causing financial loss and skewed analytics. This study presents a CNN-LSTM based click fraud detection system integrated with an interactive GUI for both training and real-time prediction. The system leverages temporal and sequential patterns in user clickstream data to identify fraudulent activity. Click features, such as click interval, session duration, clicks per session, device type, and browser type, serve as the input for a hybrid Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) architecture. CNN layers capture spatial correlations between features, while LSTM layers model temporal dependencies within sessions.The GUI allows users to upload datasets in CSV format, train the model with a standardized preprocessing pipeline, and save/load the trained model as a single bundled file using joblib. Manual input of click patterns enables real-time prediction of fraudulent behavior, providing probability scores and clear classification as "FRAUD" or "GENUINE." Additionally, the system integrates SHAP (SHapley Additive exPlanations) to provide interpretability, allowing users to understand the contribution of each feature to the predicted outcome. The experimental workflow begins with data preprocessing using StandardScaler to normalize input features. The CNN-LSTM model is trained using an 80-20 train-test split and optimized with the Adam optimizer and binary crossentropy loss. The trained model achieves high accuracy in distinguishing between fraudulent and genuine click sessions. The GUI supports iterative testing with immediate feedback, making it suitable for both research and operational deployment in ad networks.This integrated approach not only enhances detection performance but also promotes transparency and trustworthiness through explainable AI. The combination of GUI accessibility, real-time prediction, and interpretability demonstrates a practical solution for combating click fraud.