Bayesian Optimization for Parameter Selection in Fusion Systems

In this paper, we propose a methodology for the application of Bayesian Optimization to the optimization of parameters in multi-sensor fusion systems. We apply this methodology to a state-of-the-art fusion model and demonstrate its efficacy in the optimization of fusion model parameters, including temporal decay, sensor priors and the event threshold, by employing Tree-Structured Parzen Estimators. The efficacy of the proposed methodology is evaluated by comparing the performance of the optimized system with that of a standard fusion system on a data set in the context of railway security. The results demonstrate a significant improvement in key metrics such as accuracy, false positive rate and F1-Score.