Advancing Time Series Forecasting Techniques & Practices in a Big Data Environment

Accurate forecasting is pivotal to domains such as transportation, tourism, and energy. Although forecasting was traditionally limited to a few time series analysed by statisticians, the scale of the data collated has escalated rapidly in recent years. Consequently, while data scientists are becoming enthusiastic about applying Machine Learning (ML) techniques for forecasting, the details behind adapting them to forecasting remain lesser-known. This thesis addresses this overarching problem by 1) performing empirical analyses on the factors influencing the performance of ML models built as Global Forecasting Models (GFM) and 2) developing tools and guidelines to support forecast evaluation in many-series scenarios.