Browsing by Author "Ganzevoort, Reinard Christiaan"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    A decision support framework for the selection of appropriate time series forecasting methods in the retail sector
    (Stellenbosch : Stellenbosch University, 2023-10) Ganzevoort, Reinard Christiaan; Van Vuuren, Jan Harm ; Lindner, Berni G; Du Toit, Jacques; Stellenbosch University. Faculty of Engineering. Dept. of Industrial Engineering.
    ENGLISH ABSTRACT:There is a significant trade-off in any high-turnover retail environment between limiting in-store inventory levels and mitigating the risk of stock-outs. This trade-off is caused by the typical aim of retail organisations to minimise the capital tied up in inventory without incurring a significant deterioration of their service levels (i.e. to ensure product availability for customers). In order, therefore, to better manage their inventories, retailers often consider the prediction of customer behaviour as a main priority. In practice, however, sales forecasting processes are usually automated to some extent and practitioners often have limited knowledge pertaining to the selection of appropriate forecasting methods. A generic framework is proposed in this dissertation for assisting retail forecasting practitioners in the selection of appropriate forecasting methods based on available time series data sets pertaining to retail sales. This forecasting framework takes as input a multivariate time series sales data set and facilitates the configuration, transformation and extraction of valuable information from these data in order to partition the data set into clusters of time series exhibiting similar attributes. The working of the framework is based on a generic, two-phased approach. One phase of the framework, called its benchmarking phase, involves establishing a benchmark data set (or updating it if it already exists) which can be leveraged to inform feature-based forecast model identification and ranking for different clusters of time series. The computationally efficient identification of a tailored shortlist of forecast models is thus facilitated during the other framework phase, called its implementation phase, for each sales time series presented to it by a retail organisation, based on the features of the time series presented. The two phases of the framework may be applied repeatedly in alternating fashion, thus enlarging the benchmark data set and improving its representativeness each time after having applied the implementation phase to the sales time series data of a new retail organisation. The framework is verified with reference to well-established retail sales benchmark data. The verified framework is employed to evaluate the difference in forecast quality and computational time, based on the benchmark data, that results from applying the forecasting methods recommended by the framework to newly presented retail timeseries data as opposed to exhaustively applying forecasting methods classified as traditional statistical techniques, machine learning techniques and ensemble techniques. The working of the framework is finally validated by applying computerised instantiations thereof to real-world data sets of time series representing retail sales.