Study of Clustering Techniques Using Machine Learning on SET100 Stocks of the Stock Exchange of Thailand

Abstract

Securities trading is one of the investment channels that can generate substantial returns, but these returns also come with high risks. Therefore, an important responsibility of investors is to diversify risk to reduce potential losses.This article presents a method for clustering stocks in the SET100 of the Stock Exchange of Thailand using Machine Learning. It applies Unsupervised Learning techniques, including K-means, Hierarchical Clustering, DBSCAN, and Fuzzy C-means, which can group securities with similar price movements. The study collected historical data of SET 100 stocks, which consist of 100 common stocks with high market capitalization and consistent liquidity. These stocks are fundamentally strong, with continuous business growth, making them attractive to investors. For the clustering process, the study used two key factors: Dividend Yield and Stock Risk (BETA) over a three-year period, covering 274 stocks. The results from the four clustering methods showed that K-means divided the stocks into three groups: low-risk stocks with low dividends (192 stocks), high-dividend low-risk stocks (4 stocks), and growth stocks with high risk (78 stocks). Hierarchical Clustering also resulted in three groups: low-risk stocks with low dividends (163 stocks), high-dividend low-risk stocks (4 stocks), and growth stocks with high risk (107 stocks). DBSCAN identified two groups: the main cluster (266 stocks) and outlier stocks (8 stocks). Fuzzy C-means classified the stocks into three groups: high-dividend low-risk stocks (89 stocks), stocks with balanced risk and return (138 stocks), and highly volatile stocks (47 stocks).