Work

...............

Year: 2026
Type: article
Abstract: Indoor air quality (IAQ) refers to the condition of the air inside and around the building, and it directly affects the health and daily performance of people who live there. People spend 90% of their time indoors, where pollutant levels can be two to five times higher and occasionally up to 100 times higher than outdoor levels. The pollutants can be accumulated due to poor ventilation, occupancy, booking, cleaning products and outdoor infiltration. This study proposes a two stage framework for indoor air-quality and health-risk prediction using multi-pollutant sensor data. In the initial stage, Time series models are used to forecast future pollutant concentration.Environmental and pollutant variables include occupancy, temperature, humidity, PM1, PM2.5, PM10, CO2, CO, NO2, and VOC concentrations. Carbon dioxide (CO2) was selected as the primary forecasting variable because it responds strongly to occupancy and ventilation changes. In the second stage. Ensemble machine-learning models classify indoor health risk using pollutant concentrations, temperature, humidity, occupancy, and room type into high, medium and low. The dataset taken contains 25.004 indoor environment records from 5 types of rooms namely office, kitchen, classroom, living room, and bedroom. After cleaning, missing-value treatment, duplicate removal, type correction, and interquartile-range-based outlier handling, the processed data was used for exploratory analysis, forecasting, and classification. Models like Holt-Winters, ARIMA, and Prophet were trained and compared for CO₂ forecasting. The best forecasting model was chosen to be Holt-Winters producing the lowest error, with a mean absolute error of 118.231 ppm and a root mean square error of 150.053 ppm. For health-risk classification models trained and evaluated were Extra Trees, XGBoost, LightGBM, and CatBoost. LightGBM achieved the best overall classification performance, with 53.25% accuracy, 51.35% weighted precision, 53.25% weighted recall, and an F1score of 0.4990. These results indicate that combining ensemble classification with pollutant forecasting can support early awareness of dangerous indoor air conditions and help building occupants or ventilation systems respond before exposure becomes harmful.
Author: ..........
Language: English
Open Access status: green
OpenAlex ID: W7236587310
FWCI:
Cites:
Cited by:
Related to: 10