Optimized multi-class classification of PM2.5 in wooden houses at different altitudes
Abstract
This study investigates multiclass classification of PM2.5 concentrations in wooden houses located at different altitudes in Indonesia using optimized machine learning models. The dataset comprised approximately 430,000 observations from highland and lowland residential environments and included particulate, gaseous, meteorological, temporal, and location variables. Support vector machine (SVM), random forest (RF), and CatBoost were evaluated using an 80:20 stratified train-test split and five-fold cross-validation. SVM and RF were tuned using grid search, whereas CatBoost used randomized hyperparameter search with 25 candidate configurations. Optimization improved all models. RF achieved the best overall performance with 98.19% accuracy, 0.9763 macro F1-score, and 0.9655 matthews correlation coefficient (MCC), followed by CatBoost with 97.12% accuracy and SVM with 94.71% accuracy. The moderate class remained the most difficult because its feature patterns overlapped with adjacent categories. For CatBoost, optimization reduced test-set classification errors from 3,578 to 2,474 (30.9%). Shapley additive explanations (SHAP) analysis of the optimized CatBoost model using 5,000 independent test observations identified PM10 as the dominant predictor (mean |SHAP|=2.9566), followed by eCO2 and total volatile organic compounds (TVOC), while location provided a smaller but measurable contextual contribution (0.1696). The results show that ensemble models provide strong multiclass performance for heterogeneous residential PM2.5 data and that SHAP can clarify the environmental factors underlying model decisions.
Keywords
Air quality classification; CatBoost; Explainable artificial intelligence; PM2.5; Random forest; Shapley additive explanations; Support vector machine
Full Text:
PDFDOI: https://doi.org/10.11591/eei.v15i6.11705
Refbacks
- There are currently no refbacks.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.
Bulletin of Electrical Engineering and Informatics (BEEI)
ISSN: 2089-3191
,
e-ISSN: 2302-9285
This journal is published by the
Institute of Advanced Engineering and Science (IAES)
.