This research presents advanced methodologies for dynamic modeling of neural connectivity using time-series biomedical data. The framework integrates preprocessing, autoregressive and multivariate autoregressive modeling, Granger causality estimation, and time-frequency decomposition, followed by graph-theoretic metrics to capture evolving patterns of brain connectivity. The pipeline enables robust estimation of directional and frequency-specific interactions between brain regions, thereby offering improved insights into both healthy and pathological states. Experimental evaluation demonstrates superior performance across multiple metrics, including accuracy (0.92), F1-score (0.90), AUC (0.95), and MCC (0.81). The proposed approach also shows high computational efficiency (15.2 ms per inference), energy efficiency (0.42 J), and robustness under stress scenarios such as Gaussian noise, missing data, and temporal drift. These results establish the framework as both accurate and scalable, while also maintaining interpretability and resilience in real-world biomedical applications.