Explainable Machine Learning Agent for Forged Document Detection in E-Communication and Educational Services

Since even small forgeries can have major ethical and managerial ramifications, the quick development of electronic communication and web-based learning services makes document verification more difficult. Time-consuming manual checking is ineffective against complex forgeries. To close this gap, this paper proposes an Explainable Machine Learning Agent that leverages a refined ResNet50 model to identify forged documents and signatures. The proposed model uses a deep residual learning-based spatial feature extractor with SHAP-based interpretability that allows high accuracy along with transparency in decisions. The model was trained on an 80-20 stratified split of the combined dataset of handwritten signatures, namely CEDAR, BHSig260–Bengali, and BHSig260–Hindi. Key performance indicators for accuracy, precision, recall, F1-score, and AUROC were used to test the results. With an overall accuracy of 95% and an AUC of 0.96, this design surpassed previous benchmarks such VGG19, MobileNetV2, Random Forest, and Logistic Regression. In addition to performance, SHAP explainability analysis offered feature attributions that could be interpreted for forecast tracking and auditing. Therefore, the agent provides a workable, intelligible method that can be readily implemented for document forgery detection in real-world e-communications and educational verification systems.