Research on Stock Index Prediction Based on SARIMA-BP-KPCA+RF Model

The prediction of stock prices plays a vital role in guiding investors' decision - making processes and managing risks within the financial market. The stock price of ADANIPORTS is influenced by multidimensional data and exhibits nonlinear dynamic characteristics. This study is based on Yahoo Finance data, calculating 8 technical indicators including KDJ and RSI. After feature screening, seasonal autoregressive integral moving average model (SARIMA), backpropagation neural network model (BP neural network), and kernel principal component analysis random forest model (KPCA+RF) are constructed and their performance is compared. The experiment showed that the MAE of SARIMA was 8.4983 and the R² was 0.9668; The MAE of the BP neural network is 13.7540 and the R² is 0.9797; The MAE of KPCA+RF is 10.3501 and the R² is 0.9832, indicating the best prediction accuracy. The study has verified the effectiveness of ensemble learning and nonlinear dimensionality reduction, providing a multi model comparative analysis framework for stock prediction and having methodological reference value.