Prediction of e-commerce time series

OData support
Supervisor:
Dr. Pataki Béla
Department of Measurement and Information Systems

E-business companies collect a lot of data during their operation. It’s the IT professional’s role to store and analyze this huge amount of information and help the business decision makers in their work. One of the challenging tasks is to forecast the number of orders in the future. Business leaders and marketing managers are using various statistical methods for sales forecasting. Unfortunately these methods are not accurate enough. Using the modern data mining and machine learning techniques, significant enhancement can be achieved. This paper will introduce the most commonly used statistical and data mining algorithms, presenting the advantages and disadvantages of each one. At the second section my own experiments are presented. I choose LS-SVM to predict the sales of a Hungarian webshop. According to the earlier studies, LS-SVM is one of the fastest and most accurate algorithm in the family of neural networks. I used moving average as a method for comparison. I tried several different configurations to get out the most from the LS-SVM. However I had to work with a small data set, LS-SVM was able to generate prediction with less error than the traditional moving average method.

Downloads

Please sign in to download the files of this thesis.