Least-Square Support Vector Machine (LS-SVM) Parameters Optimization using Hybrid Cuckoo Search and Harmony Search Algorithm for Pre Collision Warning on Driver Assistance System (DAS)

Abstract

The growth of technology has triggered transportation sector to deliver various kind of advantages. One of them is pre collision warning. There are several steps in warning a collision, namely classifying and predicting the collisions then. Many supervised machine learning algorithms have been conducted, one of well known algorithm is Least Square – Support Vector Machine (LS-SVM). Radial Based Function (RBF) is one of LS-SVM kernels which is well-known method to support reliable performance. However, C and Gamma of its parameters are chosen randomly. This makes the performance of the classifier less optimal. So that, hybrid cuckoo search and harmony search algorithm is conducted to optimize LSSVM parameters in this research. 8437 transportation records were used in the experiment and evaluated by using accuracy. Furthermore, the proposed method was also evaluated using several metaheuristic optimization algorithms namely Cuckoo Search Algorithm (CS-SVM), Bat Algorithm (BA-SVM), and Firefly Algorithm (FA-SVM). Experimental results show that hybrid Cuckoo Search and Harmony Search Algorithm (CSHS-SVM) successfully enhance the performance by reaching 84.513% for accuracy, compared to Cuckoo Search Algorithm, Bat Algorithm, and Firefly Algorithm.