3SV-KNNC Optimization using SVR and LMKNN for Stock Price Prediction

In the capital market, a broker often finds it difficult to decide whether to buy or sell shares. Massive stock data and errors in reading stock indicators can result in a broker experiencing losses. From the abundance of stock data, it is difficult for a broker to determine rising, fixed, and falling prices because of the limited linear functions that are applied. The study purposed to divide the data into more than two data classes, namely rising prices, fixed prices, and falling prices with the regression function and predicted using classification techniques. The models proposed in this study are SV-knock-SVR (Support Vector Regression), SOM (Self Organizing Map), and LMKNN (Local Mean based K Nearest Neighbor). In this study, the data used from the Indonesia Stock Exchange with a 10-year dataset (2012 - 2021) with rising, fixed, and decreasing prices from 5 Blue Chip companies, namely Bank Rakyat Indonesia (BBRI), Astra International (ASII JK), PT. Gas Negara (PGAS), Telkom Indonesia (TLKM), and Bank Central Indonesia (BBCA). The raw data will be normalized to reduce the value of the data range. Then the indicator calculation will be carried out. Regression function used to the SVR (Support Vector Regression) to divide the data into three data classes. After that, the data is grouped and weighted on data using SOM, and the data will be classified using LMKNN (Local Mean based K Nearest Neighbor). Prediction results tested by comparing the results of the two-label SVM (Support Vector Machine) model and the three-label SVR (Support Vector Regression) model. From the test results, each of the highest accuracies are obtained in BBCV stock data 89.36% on the Laplacian kernel, ASII stock data 79.50% on the Laplacian kernel, TLKM stock data 96.43 % on the Gaussian kernel, BBRI stock data 93.53% on the Gaussian kernel and stock data PGAS 94.11% on Gaussian Kernel.