This study explores the integration of income levels and mobile app usage traffic data to enhance the predictive accuracy of machine learning models for crime classification in Lyon, France. Six models-Decision Tree, Random Forest, k-Nearest Neighbors (k-NN), Gradient Boosting Machine (GBM), Logistic Regression, and Neural Networks—were employed to predict the highest occurrences of crimes such as theft, drug use, environmental damage, scams, and assault. The research demonstrates that digital behavior patterns, particularly app traffic from platforms like Fortnite and Netflix, significantly correlate with higher crime occurrences, especially simple thefts in middle-income regions. The inclusion of app usage data alongside income levels led to substantial improvements in model performance. Neural Networks achieved the highest predictive accuracy of 98%, followed closely by Random Forest with 97%, and k-NN with 96.2%. In contrast, models relying solely on income data, such as Logistic Regression, struggled with the complexity of the dataset, achieving only 76.7% accuracy. This research underscores the importance of integrating digital behavioral data to enhance crime prediction, providing actionable insights for law enforcement, urban planners, and policymakers to develop data-driven strategies for crime prevention, resource allocation, and urban safety planning. The study also paves the way for future work incorporating real-time data streams and additional socioeconomic variables to further refine predictive models.