A COMPARATIVE ANALYSIS OF MACHINE LEARNING APPROACHES FOR SITES SUITABILITY OF BROADBAND TOWERS

Abstract

The demand for high-speed wireless communication services is increasing due to their wide applicability in daily life which requires smart planning to provide seamless coverage. Geospatial technologies play a vital role in planning by providing valuable insights into the physical and geographical aspects of a given area, but still the telecom sector does not utilize the power of geospatial machine learning approaches. The objective of the study was to (1) propose suitable sites for Base Transceiver Station (BTS) towers using Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost) and Random Forest (RF) classifier and (2) compare the models based on recall,accuracy , specificity, and Area Under the Curve (AUC).The land-use, geology, population, proximity to roads, bulk density and slope were used as exploratory variables. Models were trained on 70% data using 10-fold cross validation technique while 30% data was used for model validation using the python programming environment in ArcGIS. The results showed that the XGBoost comes with the accuracy of 97%, RF of 89%, and SVM of 57% for proposing suitable BTS sites. Relative variable importance showed that population, proximity to roads, slope, and land-use were identified as the most important exploratory variables, whereas bulk density and geology were recognized as the least relevant ones. The findings of validation matrices concluded that XGBoost is the best performing model for proposing suitable sites for BTS towers with the classification quality of 0.97%. The study reveals that these valuable insights are helpful for the telecom operators to implement XGBoost for proposing sites to increase signal strength and coverage for users.