Dal Alert!

Receive alerts from Dalhousie by text message.

X

MEC Thesis Defence - A Mahine Learning Approach to Forecasting Consumer Food Prices

Who: Jay (Jabez) Harris

Title: A Mahine Learning Approach to Forecasting Consumer Food Prices

Examining Committee:

Dr. Vlado Keselj - Faculty of Computer Science (Supervisor)
Dr. Sylvain Charlebois - Faculty of Management (Co-Supervisor)
Dr. Carolyn Watters - Faculty of Computer Science (Reader)
Dr. Vladimir Lucic - Barclays Capital (Reader)

 

Chair: Dr. Fernando Paulovich - Faculty of Computer Science

Abstract:

Building on the success of the Canada Food Price Report 2017 and its inclusion of a machine learning methodology, this thesis project posed and attempted to answer the following question, “What is the best way to predict food prices for the average Canadian consumer?” The Canada Consumer Price Index (CPI) was selected as the dependent variable and forecasted against three data models to access their predictive values. The models included the popular Holt-Winters Triple Smoothing Exponent model as a benchmark, a financial futures-market data model and a model adapted from the Canada Food Price Report 2017. The hope was to create a more robust forecast model for future Canada Food Price Reports and similar econometric predictions.

As hypothesized, the Financial Futures-Market based model outperformed the Food Price Report model with 1.6% and 2.4% average error rates respectively. Each model captured 4 of the 8 CPI food categories and apart from the CPI Seafood category in the case of the Food Price Report model, both models easily bested the popular Holt-Winters benchmark model.

The CPI Restaurant category, which was regarded as the most difficult to forecast because of its composite nature (Joutz F. L., 1997), produced the lowest error rate for both the Food Price Report and Financial Futures-Market models. Given the superb performance by both the Food Price Report and Financial Futures-Market models it is likely that even better results can be achieved by combining the datasets to share their uncommon information.

Time

Location

Room 430, Goldberg Computer Science Building