AutoMHS-GPT: Automated Model and Hyperparameter Selection with Generative Pre-Trained Model

Automated Machine Learning emerges as a solution to reduce the instantiation time of systems that rely on Artificial Intelligence (AI) by accelerating the search process for models and hyperparameters. These techniques, however, still require high execution time. In critical applications, such as intrusion detection in vehicular networks, delays in applying countermeasures can provoke accidents. Therefore, it is essential to guarantee accurate models in the shortest possible time to detect threats effectively. This work proposes AutoMHS-GPT, a system that uses generative artificial intelligence to reduce the time it takes to define hyperparameters and models when implementing machine learning to detect threats in vehicular networks. Based on a description of the problem, the generative model returns a text containing the appropriate model with its hyperparameters for training. Results show that AutoMSH-GPT produces models with higher threat classification performance than automated machine learning approaches AutoKeras and Auto-Sklearn, increasing in the best case the recall by 9%. Furthermore, the current proposal reduces the model search and training process, carrying out the task in around 30 minutes, while the other evaluated frameworks require two to three days.