Vehicular Ad Hoc Networks (VANETs) enable realtime communication among vehicles, improving traffic efficiency and road safety. However, their decentralized and highly dynamic nature exposes them to a range of misbehavior and cyberattacks such as falsified message injection, identity spoofing (Sybil), and denial-of-service (DoS). This paper consolidates the methodology and system design developed across all project phase. Using the VeReMi dataset, augmented with NS-3–simulated DDoS, Sybil, and blackhole scenarios, we build a Parallel MLP–CNN1D framework that compares a Multilayer Perceptron (MLP) with a 1D Convolutional Neural Network (CNN1D) for binary misbehavior detection. Explainability is incorporated using SHAP and LIME to interpret model decisions and highlight security-relevant features. Experimental evaluation shows that the MLP baseline reaches a test accuracy of 52.2% while the CNN1D model achieves 52.6% validation accuracy, both exceeding random (50%) performance on a highly challenging large-scale dataset of over 22 million messages. The results demonstrate that deep learning can capture meaningful attack patterns in VANET traffic, while the integrated XAI pipeline confirms that the models rely on semantically relevant kinematic and communication features rather than noise.