Fraud App Detection using Machine Learning Algorithms

The increasing number of mobile applications has resulted in a rise in malicious software, including fraud apps that take advantage of user permissions and system vulnerabilities to compromise sensitive personal and financial data. This paper introduces a comprehensive fraud detection system that leverages machine learning techniques to classify mobile applications as either benign or malicious. The system is implemented as a web-based platform using Flask, enabling users to upload, analyse, and classify APK (Android Package) files efficiently. Important features, such as application permissions and metadata (including app name, target SDK version, and file size), are extracted from APK files and transformed into feature vectors for classification by two machine learning models: an Artificial Neural Network (ANN) and a Support Vector Classifier (SVC). The ANN model achieves a classification accuracy of 92.26%, while the SVC model reaches an accuracy of 89%. To improve model performance further, a Genetic Algorithm (GA) is used for feature selection, which reduces the number of features and enhances both the computational efficiency and predictive accuracy of the models. The system offers an intuitive user interface that allows users to interact with the detection models, preview datasets, select classification algorithms, and view in-depth results, including safety recommendations for uploaded APK files. Additionally, the system provides visualizations of performance metrics and highlights the importance of specific features, improving the interpretability and transparency of the decision-making process.