A Comparative Analysis of Liquid Neural Networks and Incremental Learning Approaches for Stock Market Prediction

Due to the complexity, challenge, and fast growth of stock markets, they encourage more efficient use of financial resources and the expansion of macroeconomics. Tesla and Apple stock prices fluctuate constantly due to the impact of various industries and market conditions. The extraction of previously unknown patterns and information from time series is central to numerous real-world applications. This paper presents a comparative study of stock market prediction techniques for Tesla and Apple stocks, focusing primarily on the application of Liquid Neural Networks (LNN) and evaluating their performance against Incremental Learning methods. Employing LNN, a novel approach in the field of time-series forecasting, we developed models that leverage the dynamic and flexible structure of LNN to adapt to the complex and non-linear patterns observed in stock prices. The performance of the LNN model was quantitatively assessed using standard metrics: Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Directional Accuracy. The LNN model achieved an MSE of 0.000317, RMSE of 0.0178, MAE of 0.014131, MAPE of 1.8%, and a Directional Accuracy of 49.36%. In contrast, the Incremental Learning approach yielded an MSE of 0.54355, RMSE of 0.73725, MAE of 0.73158, MAPE of 89.39%, and Directional Accuracy of 28.37%. These results underscore the superior accuracy and effectiveness of LNNs in modeling stock market behaviors compared to traditional Incremental Learning methods. The findings not only demonstrate the potential of LNNs in financial analytics but also suggest avenues for further research into enhancing predictive accuracy and model robustness.