Arif Mudi Priyatno, Didik Dwi Prasetya, Triyanna Widiyaningtyas
Stock price prediction with stock price history and technical indicators has challenges in high feature dimensions. Recursive feature elimination feature selection is one solution to overcome this. However, there are several disadvantages to recursive feature elimination. Depending on the machine learning model used, the highest feature ranking may not always be the most relevant, and it also requires lengthy computation time. This study proposes feature selection optimization for stock price prediction using a two-stage feature selection framework. Stage 1 feature selection uses a combination of minimum and maximum that are relevant to the concept of quotient and important random forest features. Stage 2 feature selection uses recursive feature elimination. The stock data used are PT Bank Central Asia Tbk (BBCA), PT Bank Mandiri (Persero) Tbk (BMRI), and PT Bank Tabungan Negara (Persero) Tbk (BBTN). The prediction results show that the RMSE in the proposed method gets the best value under 20 features, and the MAPE value in the proposed method gets a value that is always smaller than other methods. This proves that the Two-Stage Feature Selection Framework successfully optimizes feature selection for stock price prediction, thereby reducing the error rate of stock price prediction. © 2024 IEEE.
Faculty of Engineering, Universitas Negeri Malang, Department of Electrical Engineering and Informatics, Malang, Indonesia
