Modern vehicles are vastly employing new information technologies,which provide tremendous benefits to vehicle safety and fuel economy.However,the increasing connectivity also makes the vehicle vulnerable to potentially cyberattack.The problem of ve-hicle malware detection and classification has emerged as an issue for vehicle cyber security.A malware dataset from Microsoft are ana-lyzed respect to the class frequency,classes coupling effect and feature importance.Several supervised learning methods are compared by changing the dataset volume.After that,a 3-level hierarchical method is proposed for malware classification.The first level utilizes thir-teensingle models to estimate the malware classes,which act as the input to the second level models.The second level is composed of three models,which are selected based on the performance of the first level models,while the third level model takes weighted predic-tion from the second level and generates the final malware classificationprediction.The proposed method reduces the malware classifica-tion logloss by 25.7%comparing with the best single model and is able to achieve 99.4%classification accuracy.