December 2019, University Machine Learning Project
Python Pandas Matplotlib Scipy StatsmodelsThis project was completed as a requirement for a machine learning and data science course in my final year of Software Engineering.
The goal of this project was to create a forecast for Alberta Natural Gas Pricing (AECO). Data was sourced from StatsCan and US EIA and modelled using a machine learning algorithm of our choice.
We first had to scrape the StatsCan and EIA websites for three years of data on Natural Gas production, consumption, pricing and storage volumes. After acquiring the proper data we needed to clean the data, mostly converting imperial units from the EIA website to metric.
After we created each of the datasets, we performed the Vector Auto-regression (VAR) on the first 2 years of data and compared against the most recent 9 months of data. After parameter tuning we were able to achieve an accuracy (Figure 1) which we believed was satisfactory enough to go ahead with the forecast.
Based on the data, the VAR model predicted a slight increase in 2020 AECO pricing (Figure 3), averaged over the year. Retrospectively, COVID-19 had a signifcant impact on Natural Gas supply and demand. At the time of writing, the current price of AECO is sitting at $4.62/GJ, a remarkable 200% increase from 2019 levels.