Scott, Ian JamesCorral Sobrevilla, Sergio Alfonso2025-11-072025-11-072025-10-27http://hdl.handle.net/10362/190257Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics, specialization in Data ScienceAccurate forecasting of natural gas prices play an essential role in supporting energy policy and guiding operational decisions in the context of Europe’s energy needs. As the Title Transfer Facility hub in the Netherlands serves as the most liquid and representative benchmark in the region, understanding and anticipating its price movements has become increasingly important, particularly in times of market stress. In this thesis, I develop a Hidden Markov Model to forecast day-ahead natural gas prices at the TTF hub. I use a daily dataset spanning January 2015 to February 2025, incorporating daily returns, volatility, and storage levels to capture both behavioral and fundamental drivers of price dynamics. The model achieved a Mean Absolute Percentage Error of 3.35%, with strong performance across both stable and volatile periods. As part of the evaluation, I also tested the model across multiple short-term horizons (1, 3, and 5-day forecasts) and compared it against a Naive and ARIMA baseline. Hidden Markov Model effectively identified four distinct hidden states, corresponding to different market regimes such as seasonal demand cycles, stable conditions, and periods of extreme stress. By combining regime detection with probabilistic forecasting, the model not only delivers accurate predictions but also enhances its interpretability, offering valuable insights for stakeholders in trading, policy, and infrastructure planning.engMachine LearningNatural GasTTFHidden Markov ModelsSDG 7 - Affordable and clean energySDG 8 - Decent work and economic growthSDG 9 - Industry, innovation and infrastructureSDG 11 - Sustainable cities and communitiesSDG 12 - Responsible production and consumptionSDG 13 - Climate actionmaster thesis204071771