Exploiting time series models for pricing fluctuation forecasting and its application

Abstract

This dissertation explores the use of machine learning algorithms for predicting trends in time series data, focusing on financial and weather applications. Accurate predictions are vital for making important decisions in business and engineering. The study develops new hybrid machine learning models that predict more accurately, and the presented work also studies algorithm efficiency and memory utilization. The research's core is exploring these new models that blend time series analysis, data preprocessing, and modern machine learning techniques. This combination leads to better performance in predicting complex, real-world data. A key feature of this work is how it connects these predictive models with decision-making processes and reinforcement learning. The study explores a framework for an efficient reinforcement learning-based supervisory system by establishing a relationship between predictive models and decision systems. This system is adept at guiding strategic decisions and optimizing outcomes in areas like trading and weather-related applications. The thesis details several main steps to build a robust system for forecasting complex time series data, which can help make the best decisions in trading or dealing with weather data. This work shows a new way to use machine learning for forecasting and decision-making in various fields.