Demand Forecasting in Retail : A Comparison of Time Series Analysis and Machine Learning Models

Having an accurate forecast of the upcoming demand is of utmost importance to a retail company, as it helps the retailer plan the day-to-day activities and optimize the supply chain. At the same time, retailers also gather a substantial amount of data about everything from weather conditions to promotional campaigns, having the potential to improve the forecasts when used right. The aim of this thesis is to compare time series analysis, which only utilize the past sales data, to machine learning models, which can also utilize other data, when forecasting retail sales figures. The comparison is conducted by a thorough literature review and an empirical study where forecasting is performed on a set of Walmart sales data. The forecasting methods used in the empirical study are ARIMA, Holt-Winter’s exponential smoothing, linear regression, decision tree and artificial neural networks, and the results of the empirical study suggest that ARIMA and Holt-Winter’s exponential smoothing are the best performing models on this particular dataset.