Predicting Stock Market Trends with a Transfer Learning Model Based on CNN-LSTM Architecture

Forecasting stock market trends is an essential endeavor, garnering significant interest due to the potential for substantial profits hinging on well-informed investment decisions. The endeavor of predicting stock market behavior has grown increasingly complex in recent times, largely due to the presence of disorderly data, non-stationary data patterns, and the cacophony of data sources. Consequently, this complexity presents a formidable challenge to investors seeking lucrative returns on their investments. Numerous techniques have been devised to tackle the formidable task of stock market prediction, each tailored to specific timeframes and market conditions. This research study employs a novel approach by harnessing a blended optimization algorithm to enhance the efficacy of Transfer Learning (TL) in the context of stock market prediction. Incorporating a CNN integrated with a LSTM (CNN-based LSTM), our Transfer Learning (TL) framework is adept at forecasting stock market trends by leveraging augmented data. The proposed approach utilizes time series data from three distinct companies' stock market datasets as our input. Subsequently, a multitude of features is extracted from this input dataset, and a judicious feature selection process is employed, guided by Motyka similarity metrics. To strengthen the dataset and enhance predictive accuracy, this study employs the Bootstrap method as a key component, facilitating robust data augmentation techniques. To evaluate the effectiveness of the proposed model, its performance is assessed using three distinct metrics: Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE).