Application of machine learning in the classification of area potential for transaction forecasting using regression analysis in two regions of the Johannesburg conurbation

Date
2024-03
Journal Title
Journal ISSN
Volume Title
Publisher
Stellenbosch : Stellenbosch University
Abstract
ENGLISH ABSTRACT: The optimal placement of a retail site is the most important decision a retailer can make. The “right” location should not only meet current business needs but also have long-term growth potential. Site selection can be done through consideration of the development and economic trajectory of the surrounding area. Studies on retail site selection are limited and current techniques that exist are either biased, lack quantitative measuring capability or fail to capture local or spatial variations. This study introduces a retail site selection spatial decision support system framework that coalesces the use of Earth Observation (EO) technologies, machine learning algorithms and spatial regression for the optimal placement of a retail site. The analysis was conducted in two regions of the Johannesburg conurbation. EO and machine learning algorithms were used in the classification of an economic area index that accurately depicted regions of spatial socio-economic variability. Random forest (RF) produced a marginally more accurate classification than Support Vector Machine (SVM), which was subsequently used as one of several input variables in a regression model. Spatial regression was used to forecast bank branch transactional volumes for the evaluation of optimal retail site placement. The findings revealed that study area-specific significant factors impacted the prediction of retail bank branch transactional volumes. The spatial regression detected localised spatial variations and patterns allowing for sound statistical inferences in the site selection process. The framework introduced in this study can be used to guide informed decisionmaking for the optimal placement of retail sites.
AFRIKAANSE OPSOMMING: Die optimale plasing van 'n kleinhandelsperseel is die belangrikste besluit wat 'n kleinhandelaar kan neem. Die “regte” ligging behoort nie net aan huidige sakebehoeftes te voldoen nie, maar moet ook langtermyngroeipotensiaal hê. Terreinkeuse kan gedoen word deur die ontwikkeling en ekonomiese trajek van die omliggende gebied in ag te neem. Daar is nog nie baie studies oor die keuse van kleinhandelspersele gedoen nie en die huidige tegnieke wat bestaan, is óf bevooroordeeld, gebrekkig aan kwantitatiewe meetvermoë óf versuim om plaaslike of ruimtelike variasies vas te lê. Hierdie studie stel 'n kleinhandelsterreinseleksieruimtelikebesluit- ondersteuningstelselraamwerk bekend wat die gebruik van Aardwaarneming (EO)-tegnologieë, masjienleer-algoritmes en ruimtelike regressie saamvoeg vir die optimale plasing van 'n kleinhandelsterrein. Die ontleding is in twee streke van Johannesburgse woonbuurte gedoen. EO en masjienleeralgoritmes is gebruik vir die klassifikasie van 'n ekonomiese-area-indeks wat streke van ruimtelike sosio-ekonomiese veranderlikheid akkuraat uitgebeeld het. Ewekansige-woud-masjienleer (RF) het 'n effens meer akkurate klassifikasie as ondersteuning-vektormasjien (“Support Vector Machine” (SVM)) geproduseer, wat daarná as een van verskeie insetveranderlikes in 'n regressiemodel gebruik is. Ruimtelike regressie is gebruik om banktaktransaksievolumes te voorspel vir die evaluering van optimale kleinhandelsperseelplasing. Die bevindinge het aan die lig gebring dat studie-area-spesifieke beduidende faktore die voorspelling van kleinhandelsbank-taktransaksievolumes beïnvloed het. Die ruimtelike regressie het gelokaliseerde ruimtelike variasies en patrone opgespoor wat goeie statistiese afleidings in die terreinkeuseproses moontlik gemaak het. Die raamwerk wat in hierdie studie bekendgestel is, kan gebruik word vir ingeligte besluitneming ten opsigte van die optimale plasing van kleinhandelspersele.
Description
Thesis (MA)--Stellenbosch University, 2024.
Keywords
Citation