<p>The Controller Area Network (CAN) bus serves as the primary communication channel in autonomous and connected vehicles, playing a crucial role in information exchange. Ensuring its security is vital for maintaining effective communication within the vehicle and safeguarding the safety of both drivers and passengers. Various intrusion detection techniques including those based on machine learning have been employed to secure in-vehicle CAN bus networks. Each technique has strengths and weaknesses. However, linear models have limited nonlinear capabilities and fail to account for interactions among higher-order features. As a result, they are unable to accurately represent the relationships between CAN message features and various attack modes. Therefore, a single machine learning based approach has not been able to provide complete protection against all types of real attacks. In this paper, we propose an intrusion detection system (IDS) that combines two machine learning techniques. This system is composed of a deep neural network (DNN) in the first block. Its role is to calculate the emission probabilities for the hidden Markov model (HMM) that comes as a second block. HMM will classify the packets as normal or malicious. Our model is able to efficiently recognize malicious CAN sequences using the joint training of the neural network and the postprocessor. The experimental investigation demonstrates that the proposed system gave superior results to the literature in terms of detection rates, precision, recall and F-measure.</p>