Comparative analysis of machine learning techniques in optimal site selection

Abstract

The criticality of selecting an appropriate site in business is undeniable, as a company's location plays a pivotal role in determining its performance. In recent times, the adoption of machine learning techniques for aiding in optimal site selection has gained popularity, thanks to their ability to make data-driven predictions about the suitability of different locations. In this research, we delve into the efficiency of various machine learning and Deep Learning algorithms by leveraging data from Open Street Map (OSM), WorldPop population statistics, and Bing satellite imagery. We utilized a targeted dataset from Yelp, focusing on restaurant check-ins, to gauge the rate of human interaction with businesses in various locations. Our analysis employed a range of methods, including Support Vector Regression (SVR), Random Forest, XGBoost, Ridge Regression, Lasso Regression, and ElasticNet. Additionally, we used Bing maps' satellite imagery to train different Convolutional Neural Network (CNN) models like VGG16, VGG19, ResNet, DenseNet, and InceptionV3. The efficacy of these models were evaluated using metrics such as Root Mean Squared Error (RMSE), Mean Squared Error (MSE), Mean Absolute Error (MAE), Median Absolute Error (MedAE), Max Error (ME), and Median Absolute Deviation (MAD). The selection of these algorithmic strategies was based on their successful performance in previous studies. This study also incorporated a stacking ensemble learning technique to develop a meta-model. After analyzing the error rates of these models, the most effective one for the collected data was identified. The purpose of this study is to shed light on the strengths and weaknesses of each approach and provide useful insights for those considering the integration of machine learning in the process of site selection.