Aafia Jabeen
M.Sc. Candidate
Supervisor: Dr. Ting Hu

Time Series Analysis and Forecasting with ARIMA and Prophet

Department of Computer Science
Thursday, August 1, 2018, 10:00 a.m., Room EN 2022

---

Abstract

Oil prices directly affect the global economy and has always been a major concern for both private and public sectors. Several factors such as crude volatility, weather, current and future demand and supply of oil, etc. contribute to the unpredictability and the subtle increase and decrease in prices making it notoriously difficult to forecast the oil price. With the irregular events and nonlinearity of data, the previously used statistical forecasting techniques might give us erroneous predictions. Among many approaches for data forecasting, Machine Learning forecasting models have become very popular and have been applied to major fields by data scientists in order to get better predictions and results. These models predict the future demand based on the observed trends and patterns in the data over successive time intervals.

In this project, data containing the oil prices of around 30 years has been analyzed and redictions have been made based on the observed trend in the data. It considers the Brent oil prices, one of the two grades of crude oil and also the benchmark for other oil prices. Data analysis is done using Python and Data Visualization techniques and the analyzed data is further used to train the two most popular machine learning time series forecasting models namely ARIMA and Prophet. Both the models are trained on the observed data and predictions are made on the future oil prices. The predicted data is compared to the actual data to conclude that the time series forecasting models can be used to predict future oil prices. The performance and forecasted prices of the two models are then compared to conclude that the Prophet model has a better performance and most accurate predicted oil prices when compared to ARIMA model.