Multi-Label Classification with Deep Learning for Retail Recommendation

Abstract

Selecting the right retail business for a location is crucial for the success of a business because it determines the likelihood of favourable return on investment. One common approach used in retail recommendation is multi-class classification, where retail businesses are categorized into different classes or categories based on various features or attributes. Existing research in the field of retail recommendation has extensively proposed and evaluated different algorithms, techniques, and approaches for multi-class classification in the context of retail recommendation, however, limited work has been focusing on formulating retail recommendation as a multi-label problem. This is because in retail recommendation, one location can fit multiple retail businesses so that it can provide more options to recommend the most suitable business for the location. Therefore, multi-label classification will be attempted in this study. An analytical dataset will be constructed that provides comprehensive insights into the characteristics of the business area, and subsequently employ deep learning technique for multi-label classification. The analytical dataset is constructed based on the list of sites of interest data from YellowPages, population data from Humanitarian Data Exchange (HDX) and property data sourced from brickz.my. This work will be focusing on implement deep learning technique which is 1D convolutional neural network (CNN) model. The findings showed that the proposed model achieved 61.22% in terms of accuracy.