ENHANCED STRATIFIED SAMPLING WITH REMOTE SENSING DATA FOR SOLID WASTE PROJECTS AND RESEARCH SURVEYS

Abstract

Surveys are commonly conducted for projects or research related to solid waste management. Stratified sampling allows researchers to obtain a more representative sample from a diverse population, thereby enhancing accuracy and precision, increasing efficiency, and facilitating comparisons between groups. Remote sensing product data refers to information acquired through remote sensing technologies, including socio-economic conditions. In this study, remote sensing product data were utilized to develop an index or classification of areas potentially generating waste and plastic waste. A sample of 400 households was collected, divided into two sampling approaches: 200 households were selected through stratified sampling based on the remote sensing data indexing, while 200 households were chosen randomly. Data distribution analysis was conducted using descriptive statistics, box-plot analysis, and histograms. The data distribution from stratified sampling tends to be more normal, as indicated by a histogram that is symmetrical and bell-shaped (normal distribution) and by a lower number of outliers compared to random sampling. Additionally, hypothesis testing was conducted using t-tests to examine significant differences between the two sampling methods, revealing a significant difference in waste and plastic waste generation data. Modeling using Linear Regression, Random Forest, and XGBoost was conducted, yielding MSE, RMSE, and MAE results. The stratified sampling was better (MSE: 0.13, 0.01, and 0.00, respectively) compared to random sampling (MSE: 0.23, 0.04, and 0.00, respectively). The results show that stratified random sampling, with class divisions based on remote sensing product data, yields more normalized data, less error, and greater stability for modeling analysis.