Article: Stock prediction and selection method based on LSTM-BPNN and multi-factor quantisation Journal: International Journal of Reasoning-based Intelligent Systems (IJRIS) 2026 Vol.18 No.12 pp.1 - 17 Abstract: To overcome the limitations of traditional models in capturing temporal characteristics and fitting nonlinear relationships in stock prediction and selection, and to improve prediction accuracy and return risk balance, LSTM-BPNN stock trend prediction model and PCA-BPNN multi factor stock selection model were developed. The experiment showed that the prediction model test set MAE was 0.287 CNY, RMSE was 0.392 CNY, MAPE 1.76%, significantly lower than ARIMA, and stable under different market conditions; the stock selection model extracts 6 principal components (with a cumulative variance contribution rate of 87.28%), resulting in an annualised return rate of 18.7% and a cumulative return rate of 64.3% for the portfolio from 2022 to 2024. The Sharpe ratio is 1.62, better than benchmarks such as the Shanghai and Shenzhen 300 Index. The two models provide new methods for quantitative stock analysis and assist in investment decision-making. Inderscience Publishers - linking academia, business and industry through research

Title: Stock prediction and selection method based on LSTM-BPNN and multi-factor quantisation

Authors: Jie Deng

Addresses: School of Financial Engineering and Modern Technology, Shaanxi Technical College of Finance and Economics, Xianyang, 712000, China

Abstract: To overcome the limitations of traditional models in capturing temporal characteristics and fitting nonlinear relationships in stock prediction and selection, and to improve prediction accuracy and return risk balance, LSTM-BPNN stock trend prediction model and PCA-BPNN multi factor stock selection model were developed. The experiment showed that the prediction model test set MAE was 0.287 CNY, RMSE was 0.392 CNY, MAPE 1.76%, significantly lower than ARIMA, and stable under different market conditions; the stock selection model extracts 6 principal components (with a cumulative variance contribution rate of 87.28%), resulting in an annualised return rate of 18.7% and a cumulative return rate of 64.3% for the portfolio from 2022 to 2024. The Sharpe ratio is 1.62, better than benchmarks such as the Shanghai and Shenzhen 300 Index. The two models provide new methods for quantitative stock analysis and assist in investment decision-making.

Keywords: BP neural network; LSTM; PCA; multi-factor quantification; stock prediction; stock selection strategy.

DOI: 10.1504/IJRIS.2026.153459

International Journal of Reasoning-based Intelligent Systems, 2026 Vol.18 No.12, pp.1 - 17

Received: 23 Dec 2025
Accepted: 02 Mar 2026
Published online: 08 May 2026 *

Title: Stock prediction and selection method based on LSTM-BPNN and multi-factor quantisation

Keep up-to-date