| Title |
A Bayesian Network-based Machine-learning Algorithm for Enhancing Performance in Typical Scenarios |
| DOI |
https://doi.org/10.5573/IEIESPC.2023.12.3.252 |
| Keywords |
Machine-learning; Bayesian; Automation; CASH; Artificial intelligence |
| Abstract |
Automated machine-learning technology can achieve the automation, efficiency, and intelligence of machine learning. This technology can lower the application threshold of artificial intelligence (AI) and has attracted academic attention. Therefore, taking a typical classification problem as an example, this study proposes a framework of a machine-learning pipeline automation design algorithm combining a Bayesian algorithm and reinforcement learning. Aiming at the CASH problem, the study divides the machine-learning pipeline design problem into two sub-problems. One is to realize a machine-learning pipeline structure search based on reinforcement learning. The other is to realize machine-learning pipeline hyperparameter optimization configurations based on the Bayesian network model. The experimental results showed that when the time budget was four hours, the average balanced accuracy of Auto-PLD(Ed note: You need to define what these are in the abstract.) was 0.003 higher than that of Auto-PLD-random and 0.001 higher than that of Auto-sklearn. The success rate of the Auto-PLD machine-learning pipeline evaluation on various datasets exceeded 92%. Based on the Bayesian model and reinforcement learning, the machine-learning pipeline automation design algorithm framework can also play a good role in practical applications. Moreover, it can promote the development of artificial intelligence technology. |