Back to results list
Please use this identifier to cite or link to this item:
|Title:||Panel data based fashion sales forecasting systems||Authors:||Ren, Shuyun||Advisors:||Choi, Jason T. M. (ITC)||Keywords:||Fashion -- Forecasting.
Clothing trade -- Forecasting.
|Issue Date:||2016||Publisher:||The Hong Kong Polytechnic University||Abstract:||Today's fashion industry is known to be information driven. The wise use of information for conducting sales forecasting helps a lot in enhancing the operations management of fashion companies. Numerous forecasting methods, such as statistical methods, Artificial Intelligence (AI) methods and many different kinds of hybrid methods, have been developed and studied in the literature for over two decades.However,good fashion sales forecasting is very difficult to achieve because the short product life-cycle and the ever-changing fashion trend make the demand of fashion product highly volatile.In this thesis, to analyze and compare the cutting edge technologies of fashion sales forecasting, a comprehensive literature review was first conducted. It was found that pure one-dimensional time series based statistical methods were easy to implement and able to provide analytical solutions, but failed to yield a top forecasting performance. AI models were powerful for fashion sales forecasting, but they were computationally time consuming and usually required a large amount of historical data. Thus, in this thesis, panel data methods were introduced to conduct fashion sales forecasting. Panel data methods were widely used to do sales forecasting in various industrial settings. It could model the influence from other correlated products and some important related factors that pure time-series based single-dimension statistical methods missed through its multi-dimensional data structure.
Useful fashion sales forecasting models should be applications oriented and hence industrialists' feelings and criteria should be examined. In order to reveal some insights regarding how Industrialists evaluated different major sales forecasting methods, an industrial survey was conducted with an aim to examine the industrialists' preference on fashion sales forecasting models. With the collected data, an analytic hierarchy process (AHP) analysis was conducted. To further investigate the preference of different decision makers with different roles, comparison studies were conducted by filtering the survey data into three category groups. Some important findings, including the usefulness of panel data based models, were obtained. After that, in order to develop a versatile and innovative fashion sales forecasting application, a panel data based particle filter (PDPF) model was proposed for conducting fashion sales forecasting. The core advantage of this hybrid PDPF model was its three-dimensional correlation structure which could incorporate the influence of the previous sales of the specific product item, its price, and the effects brought by other correlated product items, into the forecasting model. A computational analysis, using real sales data, was conducted to further examine the forecasting performance of the proposed PDPF (versus other commonly seen methods reported in the literature). It was found that the PDPF outperformed the other methods. Moreover, some important relationships, such as 1) the relationship between sales and corresponding price, 2) the relationship between the amount of historical data and the forecasting performance and 3) the relationship between the frequency of information updating and forecasting performance were all investigated. Important insights were generated.
|Description:||PolyU Library Call No.: [THS] LG51 .H577P ITC 2016 Ren
vii, 137 pages :color illustrations
|URI:||http://hdl.handle.net/10397/55263||Rights:||All rights reserved.|
|Appears in Collections:||Thesis|
Show full item record
Files in This Item:
|b28945463_link.htm||For PolyU Users||208 B||HTML||View/Open|
|b28945463_ira.pdf||For All Users (Non-printable)||1.33 MB||Adobe PDF||View/Open|
Citations as of May 21, 2018
Citations as of May 21, 2018
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.