TransDenseInceptionNet: A Deep Learning Framework for Teenage Cybersecurity Awareness Using Real-World E-Safety Data

The rapid rise in teenage internet use has heightened the need for effective e-safety and cybersecurity measures. However, existing models often lack the precision and adaptability required to address the complex and evolving patterns of teenage online activity. This study proposes TransDenseInceptionNet (TDINet), a hybrid deep learning model that integrates DenseNet for feature reuse, InceptionNet for multi-scale feature extraction, and Transformer layers for long-range interactions. The model is trained on a longitudinal dataset (2017-2024) from Texas and California, which includes key cybersecurity indicators such as device types, social media usage, malware detection, password strength, and security incidents. To address data imbalances, outliers, and complex feature interactions, we introduce a robust preprocessing pipeline incorporating Dynamic Feature Imbalance Compensation (DFIC), Cumulative Anomaly Weighting (CAW), and Adaptive Projection Encoding (APE). Additionally, Contextual Feature Synthesis (CFS) enhances prediction accuracy by capturing intricate interaction patterns. Simulations conducted using TensorFlow GPU in Google Colab demonstrate that TDINet achieves 97% accuracy, 0.99 AUC, 96.5% F1-score, and superior performance in precision (96.8%) and recall (97.1%) compared to CNN, LSTM, and GNN models. The novel preprocessing techniques improve feature representation, leading to more robust and stable learning. Furthermore, novel evaluation metrics, including Adaptive Interaction Efficiency (AIE), Temporal Stability Index (TSI), and Anomaly Sensitivity Factor (ASF), validate TDINet’s reliability in detecting anomalies with low false positive rates and maintaining prediction stability. The results underscore that TDINet analyzes historical data to classify behaviors and forecast cybersecurity risks based on learned trends, offering a scalable and impactful solution for improved cybersecurity in adolescent online behavior.