Optimized multi-output machine learning system for engineering informatics in assessing natural hazards

Abstract

This work develops a novel metaheuristic optimization-based least squares support vector regression (LSSVR) model with a multi-output (MO) algorithm for assessing natural hazards. The MO algorithm is more efficient than the single-output algorithm because the relations among outputs can be estimated simultaneously by the proposed prediction model. Furthermore, the hyperparameters in MOLSSVR are optimized using an accelerated particle swarm optimization (APSO) algorithm combined with a self-tuning method to generate the best predictions and the fastest convergence. The APSO algorithm is validated by solving benchmark functions with unimodal and multimodal characteristics. The performance of APSO-MOLSSVR is compared with those of hybrid and single models yielded from standard multi-input single-output algorithms. A graphical user interface was designed as a stand-alone application to provide a user-friendly system for executing advanced data mining techniques. In real-world engineering cases, APSO-MOLSSVR achieved an error rate that was up to 63.55% better than those achieved using prediction models that are proposed in the single-output scheme. The system much more quickly and efficiently identified the optimal parameters and effectively solved multiple-output problems.