By combining graph-theoretic modelling and artificial intelligence methods, this study shows a new way to find side-channel attacks.  Data from the side channels is shown as graphs that show the connections and patterns that could be used in an attack.  Feature extraction techniques are used on these graph models to show off unique attack patterns.  Graph Neural Networks (GNNs) and Support Vector Machines (SVMs) are two advanced AI algorithms that are designed and trained to accurately find side-channel intrusions.  Discrete mathematics ideas like vector measure spaces and orthogonal sequences are used in the suggested method to make identification more reliable.  Evaluation measures and validation methods are used to check the accuracy and dependability of detection, showing that the framework works well at finding both simple and complex side-channel dangers.