Abstract: This study presents the optimum financial index classification model that minimizes the pattern classification risk by designing and comparing various statistical and artificial intelligence models for classifying KOSPI200 (Korea Composite Stock Price Index 200) patterns. The models used for the experiment include Logistic Regression Analysis, Discriminant Analysis, Neural Network, and the Support Vector Machine which is in the latest spotlight. The simulation was performed by designing the models that create the binary classification of the next day pattern (up/down) from the daily closing value of KOSPI200 and then comparing their performance to present the best financial index pattern classification model. The Neural Network used back propagation algorithm, and the kernel for SVM used the polynomial and RBF functions. T-test and then the logistic regression analysis were performed to select the variables in two phases. The simulation showed that SVM with RBF kernel function resulted in the lowest misclassification rate of 0.36%.
, 2004. Design of the Data Mining Classification Model and Their Performance Comparison for Financial Index, KOSPI200 - SVM, Neural Network, Logistic Regression Analysis, Discriminant Analysis . Asian Journal of Information Technology, 3: 1259-1270.