Cyber-physical system (CPS) of EV on-board chargers is connected to an IOT-based communication network for coordinated control, which is highly vulnerable to cyber-attacks. This charging coordination control incorporating hundreds of EVs and associated charging sessions, feed in a stochastic reference input to energy management system (EMS) of on-board EV chargers. Hence, under these varying operating conditions, a pure data-driven-based detection model can experience a disturbance detection failure. Therefore, a model predictive control (MPC) based machine learning (ML) network, integrated with a residual based training data pre-processing is proposed in this paper. This MPC based ML approach can effectively detect a tempered response while addressing the aleatory behaviour of cooperative control with enhanced disturbance detection accuracy. The proposed model utilizing various system level signals can also efficiently classify a normal condition, cyber-attack, and a physical fault. The superior performance of the proposed approach is validated by using different case study scenarios of training datasets.