Machine Learning-Based Detection of Malicious Applications in App Stores

Abstract

With the exponential growth of mobile applications, users increasingly rely on app stores for essential services, making them vulnerable to fraudulent and malicious applications. Detecting such apps manually is difficult due to the vast number of available applications and misleading ratings or descriptions. To address this challenge, this project presents a machine learning–based approach for detecting fraudulent applications using sentiment analysis of user reviews. The proposed system analyzes textual user feedback to classify sentiments as positive, negative, or neutral, helping to uncover hidden patterns of suspicious behaviour. In addition to sentiment scores, features such as app ratings, review consistency, and app metadata are considered to improve detection accuracy. Multiple classification models, including Naïve Bayes, Logistic Regression, and Decision Tree Classifier, are implemented and evaluated. The performance of these models is measured using accuracy, precision, recall, and F1-score to ensure reliable classification. Experimental results demonstrate that sentiment analysis significantly enhances fraud detection capability by capturing real user experiences. Among the evaluated models, the Decision Tree Classifier achieved superior performance with high accuracy and F1-score, making it the most effective approach for identifying potentially fraudulent applications. This system provides an efficient and scalable solution to improve user trust and ensure a safer mobile app ecosystem.