Stock Trend Forecasting Using Hybrid Particle Swarm Optimization-Support Vector Machine (PSOSVM) Technique

Lee, Zhong Zhen (2011) Stock Trend Forecasting Using Hybrid Particle Swarm Optimization-Support Vector Machine (PSOSVM) Technique. Project Report. UTeM, Melaka. (Submitted)

	PDF (24 Pages) Stock_Trend_Forecasting_Using_Hybrid_Particle_Swarm_Optimization-Support_Vector_Machine_(PSOSVM)_Technique_24_Pages.pdf - Submitted Version Download (3MB)
	PDF (Full Text) Stock_Trend_Forecasting_Using_Hybrid_Particle_Swarm_Optimization-Support_Vector_Machine_(PSOSVM)_Technique_Full_Text.pdf - Submitted Version Restricted to Registered users only Download (14MB)

Abstract

Stock trend forecasting is one of the important issues in stock market research. Accurate stock trend prediction by using a well suited algorithm is a tough challenge in financial industries because the distribution of stock data differs over time. In such case, Support Vector Machine (SVM) produces a fairly good result in stock trend forecasting. However, the accuracy of SVM can be affected if there exists too many input features or the data are noisy. Hence, a feature selection technique is hybridized with SVM in order to improve the forecasting accuracy in Malaysian stock market price trend. Then, the prediction results produced will be analysed. The Improvement Concept Selection Methodology (ICSM) is used as the methodology for this research project since ICSM is designed for R&D environments. A hybrid Particle Swarm Optimization - Support Vector Machine (PSOSVM) algorithm is implemented in stock trend analysis to improve the prediction results. This project implemented SVM with RBF kernel function and optimized two parameters, i.e. they and large margin parameter automatically using PSO method. The PSOSVM algorithm was tested on pre-sampled 17 years record of daily KLCI data. The experimental results show that PSOSVM has outperformed SVM technique significantly. Thus, it is concluded that the PSOSVM technique can be implemented in forecasting systems to improve prediction accuracy. It is suggested that future work extended from this project may focus on the selection of different particle in PSO as well as different kernel function in SVM.

Actions (login required)

View Item