Abstract
The forecasting problems have evoked a lot of research interests in the past. Recently, studies have demonstrated that machine learning techniques can achieve better performance than tradi- tional statistical methods. The support vector machine (SVM) is one kind of machine learning techniques with good forecasting accuracy when with optimal kernel parameters. The SVMs are a set of related supervised learning methods that analyze data and recognize patterns, used for clas- sification and regression analysis. In past research, SVM had its higher computational burden for the constrained optimization programming and least squares support vector machine (LS-SVM) solved linear equations instead of a quadratic programming problem. Therefore, comparison of SVM, LS-SVM have smaller computational burden. Due to the importance of parameters opti- mization in LS-SVM model, the real valued genetic algorithm (RGA) and the particle swarm op- timization (PSO) were used to optimize the kernel parameter and tradeoff parameter in LS-SVM. The RGA uses a real value as a parameter of the chromosome in populations to reduce the comput- ing process. The PSO is a stochastic optimization technique which is inspired by social behavior of bird flocking and fish schooling. To determine the optimal parameters in a LS-SVM, this study proposed two novel evolutionary algorithms, the RGA and the PSO based LS-SVM, for enhancing forecasting accuracy and the efficiency of obtaining the kernel parameters. The two methods will further be verified by predicting the market trends of communication industries. The forecasts of three market trends in the communication industry are forecasting in the empirical study. The pro- posed RGA based LS-SVM and PSO based LS-SVM were verified by using the mobile phone and broadband technology market growth forecasting. The forecast efficiency is satisfactory; all fore- casting errors in the empirical study results can be compressed to fewer than 5%. In the future, the GA and PSO based LS-SVM can further be applied on other forecasts of industry market growth or trends.