Authenticity Accuracy Improvement Through the Analysis of Signature Ownership Using Convolutional Neural Network Algorithm

This research aims to improve the accuracy of signature authenticity classification using a Convolutional Neural Network (CNN) model, implemented in a web-based application using the Flask framework. In the digital era, signature authentication has become a crucial component in maintaining data security and transaction validity. However, the classification of genuine and forged signatures presents its own challenges due to the unique variations in patterns and styles of each individual. Using a public dataset from Kaggle consisting of 1,084 signature images (620 forged and 464 genuine), the CNN model was trained to recognize important patterns that can differentiate genuine signatures from forged ones. The research stages include data preprocessing, CNN model training, and evaluation using Confusion Matrix metrics, including precision, recall, and F1-score, to ensure the accuracy of prediction results. The results show that the implemented CNN model achieved an accuracy of 98% in signature classification, proving its effectiveness in distinguishing between genuine and forged signatures. Additionally, the integration of the model into a Flask-based application allows users to upload signature images and receive real-time classification results, enhancing user convenience and practicality. In conclusion, this CNN model can serve as a reliable signature-based authentication solution and has the potential to be applied in various digital security applications. This research contributes to the development of more advanced and secure digital signature authentication systems.