Robust optimization of SVM hyper-parameters for spillway type selection

Abstract

Spillways, which play a vital role in dams, can be built in various types. Although several studies have been conducted on hydraulic calculations of spillways, studies on type selection that require heuristics knowledge were limited. The tuning of the hyperparameters in machine learning algorithms is still an open problem. In this paper, a parallel global optimization algorithm is proposed optimizing the hyperparameters of a Support Vector Machine (SVM) classification model for providing accurate spillway type selection (STS). The random forest method is used to obtain the relative importance of input variables. Besides, a novel spillway dataset was introduced and a novel STS software tool has been developed based on different machine learning algorithms. Several experiments are carried out to demonstrate the effectiveness of the proposed tool and the reliability of data. The hyper-parameters optimized SVM was achieved the best results with 93.81% classification accuracy.