Signature fraud is necessary to confirm the validity of handwritten signature in particularly economic and legal sectors. Traditional techniques for identifying false signatures often show deficiencies in strength and accuracy, resulting in risk. This study suggests a deep learning method for detecting signed fraud through sophisticated Convolutional Neural Networks (CNN). Our model uses the ResNet50 for extraction of functions, and guarantees great accuracy in differences between authentic and false signatures. We use model checkpoint and Early Stopping to adapt the performance to the model, reduce overfitting and improve generalization. K-Fold cross-validation has been used to increase the strength of models in different data department, so credibility increases. Experimental findings suggest that our function improves fraud detection compared to traditional techniques. This deep learning -based system provides a scalable and effective solution for automated signature verification, which provides it particularly relevant in economic, legal and security sectors.