Improving Stock Market Trend Forecasting Using ARIMA and Machine Learning

Predicting stock market trends accurately is challenging due to the natural unpredictability and complexity of financial data. Traditional time series models, such as ARIMA, are effective in identifying linear patterns but struggle to model the more complex and nonlinear behavior often seen in financial markets. Conversely, machine learning methods like Random Forest (RF) classifiers are proficient at capturing complex relationships but often face challenges when it comes to effectively modeling time-dependent patterns. In this study, we investigate a technique for predicting stock market trends by integrating traditional machine learning models with time series analysis. Using stock data from ICICI Bank and SBI Bank from January 2014 to August 2024, we classify market movements as either bullish or bearish based on the relationship between closing and opening prices. Initially, we used default stock market features like ‘Open’, 'Close ‘, ‘Low’, ‘High’ and ‘Volume’ to train various classical ML models, such as Logistic Regression (LR), SVM, Naive Bayes (NB), k-NN,, Gradient Boosting, CatBoost, LightGBM, and XGBoost. Later, we added technical indicators like RSI, SMA, and Stochastic_K to improve the accuracy. Finally, we introduced a hybrid approach by integrating ARIMA to model the time-series behavior of stock prices, using ARIMA residuals as additional features for the models. Our results show that combining ARIMA with machine learning improves the accuracy of stock trend predictions, offering a better way to forecast financial markets.