Projects

A Comparison Analysis of Accuracy and Reliability Rates in Stock Market Forecasting Models

Programming Language/IDE: Python, Jupyter Notebook

Description: With an increasing interest in artificial intelligence and machine learning developing in the past decade, the financial sector has turned its attention to stock market forecasting due to its predictive qualities and potential for maximizing returns. Investors often rely on methods such as ARIMA, KNN and LSTM to get a sense of the direction of the market, so they can determine their next course of action. However, because these methods are relatively new, they may not be as accurate or reliable as needed for this particular financial activity. This is due to many factors, including the market being affected by external and unpredictable forces, usually as a result of human behavior. Therefore, when evaluating these predictive models for accuracy and reliability, we can adjust the flaws in their forecasting and create improved methods for predicting stock prices. After running through a detailed structural analysis, including preprocessing, training, testing, modeling and fitting, forecasting, then calculating error values, we have reached the conclusion that ARIMA stands out as the most accurate method, KNN as most reliable, and LSTM as the least accurate and reliable of the three. However, the limitations of LSTM are well known and since the model incorporates a neural network, it needs more data to learn through several cycles and cannot depend on recent observations to attain greater accuracy and reliability. 

Keywords: Machine learning, ARIMA, LSTM, K-Nearest Neighbors (KNN), Stock Market, Time Series, Stock Market Forecasting, Predictive Modeling