Deep Learning for Financial Forecasting: Evaluating CNN and CNN-LSTM in Indian Stock Market Prediction

Abstract

Stock market forecasting is a complex yet crucial task in financial analytics, as accurate predictions can significantly aid investors, analysts, and policymakers in making informed decisions. Traditional statistical models such as ARIMA have long been employed for time-series forecasting; however, they often struggle with capturing the non-linearity and long-term dependencies inherent in financial data. In recent years, deep learning architectures have demonstrated remarkable advancements in stock price prediction, particularly Long Short-Term Memory (LSTM) networks for sequential modeling and Convolutional Neural Networks (CNNs) for feature extraction. While CNNs excel in capturing local patterns in financial time series, LSTMs are highly effective in learning long-term dependencies. However, limited research has systematically compared these architectures, especially in the context of the Indian stock market, where market conditions are influenced by macroeconomic factors, sectoral trends, and trading behavior unique to emerging economies. This study aims to conduct a comparative analysis of CNN and CNN-LSTM hybrid models for stock price prediction using historical data from the Indian stock market. The models are trained on selected NIFTY50 constituents which have been part of the index since its inception. The research evaluates model performance by examining predictive accuracy, computational efficiency, and adaptability to fluctuating trends. Our findings indicate that while CNNs perform well in short-term trend analysis, CNN-LSTM models demonstrate superior robustness in capturing long-range dependencies, making them more effective for medium- to long-term forecasting in financial markets.