E-COMMERCE CUSTOMER BEHAVIOR PREDICTION USING ONLINE GAUSSIAN PROCESS

• Main Author: Anastasia, Renata
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/38978
• Institution: Institut Teknologi Bandung

Development of e-commerce companies in Indonesia drives the use of machine learning to predict customer behavior in order to increase sales. Ordinary Gaussian Process can be used to predict customer behavior since it is a non-parametric method and can be used for both regression and classification predictions. Customer behavior prediction is done in a certain period with data that comes continuously (streaming data). Therefore, training data will increase over time and Ordinary Gaussian Process will require large memory and large computational time. One of the methods to tackle this problem is too use Online Gaussian Process where variables for each iteration are calculated using variables from the previous iteration. Online Gaussian Process will also use sparse approximation and a constraint of maximum number of elements in vector base to reduce required memory and computing time. In this paper, Online Gaussian Process shows good performance in predicting regression and classification in several generated data. Online Gaussian Process is also used to solve e-commerce digital advertising problem and recommendation engines for e-commerce customer which resulted in good predictions.