Time Series Analysis

Abstract

This master’s thesis is focusing on time series forecasting, a widely studied field in data science and statistics. The aim of this thesis is to gain insight into forecasting methods, test them on data, compare them, and provide visualizations. The first chapter defines multiple statistical tools, data preprocessing techniques, evaluation metrics, and more, providing a deeper understanding of forecasting from a mathematical perspective. The following chapters focus on statistical, machine learning, and neural network approaches for forecasting. My primary focus is on visualizing each method and providing a fundamental description of their principles. This thesis also addresses multiple problems that may arise during the forecasting process, describing their causes and possible solutions. All of this information aims to contribute to an effective understanding of forecasting methods.In the final chapters, a comparison of each method is presented on multiple data types where each method may have some advantage or disadvantage. The thesis includes visual results for forecasting, as well as concrete values resulting from evaluation metrics. The conclusion summarizes the main findings and contributions of the thesis.