03464nas a2200277 a 450000500170000000800410001703500150005810000380007324500960011126000670020730000430027449000270031750001500034450200590049452021860055365000310273965000480277065000210281870000380283970000430287770000450292071000570296571000400302281000590306285600650312120260817163644.0251103s20259999th u  ms t    000   eng d  a.b124670421 aTesfamarian, Abraham Habtemichael10aAssessment of solid waste generation and compositions in AIT :ba machine learning approach  aPathum Thani, Thailand :bAsian Institute of Technology,c2025  a105 leaves :bill.+e1 online resource1 aThesis ;vno. EV-25-01  aA thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Environmental Engineering and Management  aThesis (M. Eng.) - Asian Institute of Technology, 2025  aThis study explored the generation and composition of solid waste at the Asian Institute of  Technology (AIT) in Thailand by applying an integrated approach that combines empirical  waste analysis with machine learning (ML) techniques. Data was gathered through field  sampling, online surveys, interviews, institutional sources (OFAM and OSA), and prior  thesis work. Waste composition was examined using the quartering method across  academic, residential, and commercial areas, while recyclable potential was analyzed to  assess opportunities for resource recovery.  A seasonal study (Dec 2024{u2013}Mar 2025) showed that food waste made up the largest  portion (51.24%), followed by plastics (17.58%). In the current year, AIT generated 2.08  tons/day, averaging 0.8 kg/person. In major events (e.g., graduation), waste level exceeded  80 kg/collector/trip, 5 times the baseline, but contributed only 8% of the total annual  volume. Regular weekdays cumulatively contributed 92%, suggesting targeted  interventions. Per capita waste generation rose to 0.9 kg/day during COVID-19 but  stabilized post-pandemic. Eight ML models were tested: regression (RF, SVR, GB, XGBoost) and classification (RF,  J48, LR, XGBoost). Models used an 80/20 training-test split and were evaluated using  five-fold cross-validation. SVR (R²=0.90, MAE=0.06) and RF (accuracy=82%)  outperformed others. SHAP analysis revealed that kitchen access, household size, and age  range were significant factors influencing waste generation. This research improved upon D. Zhang et al. (2020) by using zone-wide quartering instead  of single-zone sampling. Also advancing Cha et al. (2023), by combining SHAP values  with heat maps, pairing SVR with RF, and reducing MSE by 90.3% (from 0.31 to 0.03).  The R² was also 45% higher than Cha{u2019}s AE-ANN model.The framework supports SDGs 11 and 12, showing potential to reduce landfill dependence  by 35%, cut paper waste by 50% through two-sided printing and an electronic mailing policy, and divert 72% of event waste via targeted segregation. The findings offer  interpretable, ML-based campus waste solutions, potentially scalable to other institutions.  0aWaste managementzThailand 0aIntegrated solid waste managementzThailand 0aMachine learning0 aThammarat Koottatep,eChairperson1 aGhimire, Anish,eExamination Committee0 aSarawut Ninsawat,eExamination Committee2 aMinistry of Agriculture, Eritrea,eScholarship Donor2 aAIT Scholarship,eScholarship Donor2 aAsian Institute of Technology.tThesis ;vno. EV-25-01403Full-Textuhttp://203.159.5.9/ait-thesis/detail.php?q=B22989