The rapid expansion of Android devices in urban environments has led to an increase in sophisticated cyber threats, including zero-day malware, which pose risks to public safety and critical infrastructure. Traditional signature-based detection methods are often ineffective against these evolving threats. This paper presents a multi-task learning (MTL) framework designed to enhance Android malware detection and classification, contributing to the resilience of urban digital ecosystems. The model simultaneously performs binary classification (malware detection) and multi-class classification (malware family identification), using shared representations to improve efficiency and accuracy compared to single-task learning (STL) models. The framework is trained and evaluated on the CCCS-CIC-AndMal-2020 dataset, utilizing API-based static features of Android applications. Feature dimensionality is reduced through Principal Component Analysis (PCA), and class imbalance is addressed with a weighted loss function. Hyperparameter tuning with Optuna further optimizes configurations, including layer sizes, learning rate, and task-specific weights. Experimental results demonstrate that the MTL framework outperforms STL models, offering a promising approach for strengthening cybersecurity. Future work will explore real-time deployment and dynamic analysis to enhance urban resilience against emerging cyber threats.