Open AccessRetail Site Selection using Machine Learning Algorithms
Author(s) -
Hui-Jia Yee*,
Choo-Yee Ting,
Chiung Ching Ho
Publication year - 2019
Publication title -
international journal of recent technology and engineering
Language(s) - English
Resource type - Journals
ISSN - 2277-3878
DOI - 10.35940/ijrte.d7186.118419
Subject(s) - random forest , feature selection , machine learning , computer science , geospatial analysis , boosting (machine learning) , artificial intelligence , data mining , classifier (uml) , gradient boosting , support vector machine , site selection , multiple criteria decision analysis , operations research , engineering , geography , cartography , political science , law
Selecting a new site for retail business expansion has always been a challenge for decision-makers. It requires not only the sales data but the geographic data in order to decide the potential location for their respective purposes. Proper use of the data could lead to better decision-making. To date, common techniques such as geographic information system (GIS) and multi-criteria decision making (MCDM) have been applied to site selection. These methods, however, require not only extensive human effort but more importantly, difficult to validate the importance of identified variables. In this work, sales performance is proposed as a function of geospatial features to determine the suitability of a retail location. The main aim of this study was to identify features attributed to optimal site selection which in turn facilitate sales prediction for a telecommunication company in Malaysia. In this research, various feature selection techniques and machine learning models were deployed for sales prediction in order to determine the suitability of the new location. The findings show the top 3 feature selections are prediction step in VSURF, random search, and fuse learner with search strategy; the top 3 families are boosting, random forest and bagging; and the top 3 classifiers are C5.0, rf, and parRF. The crossover combination of the top feature selection-classifier can produce the AUC of more than 0.75. The highest AUC, 0.8354 was obtained through random search-parRF.