Plant Impact on Indoor Carbon Dioxide Concentration Using Ensemble Voting Prediction Models

Abstract

Indoor air quality, particularly carbon dioxide (CO₂) levels, is critical to occupants’ health and comfort. This study developed predictive models for indoor CO₂ concentrations based on environmental variables, including light, temperature, humidity, and the presence of plants. Data collected from sensors within a controlled indoor environment were used to train predictive models using various techniques, including Artificial Neural Networks (ANN), k-Nearest Neighbors (k-NN), Random Forest, and Generalized Linear Models. Among standalone models, the ANN with a 70:30 train-test split yielded the best performance, achieving a root mean square error (RMSE) of 10.960, mean absolute error (MAE) of 7.300, and a coefficient of determination (R²) of 0.640. The study further explored ensemble methods by combining ANN, k-NN, and Generalized Linear Models through soft voting. The optimal ensemble configuration—ANN and k-NN with a 90:10 split ratio—achieved an RMSE of 11.437, MAE of 8.153, and R² of 0.650, outperforming the standalone models. In addition, the results demonstrated that the presence of plants within a room reduced CO₂ levels under specific conditions (20-30°C and 200 lux), highlighting plants’ potential to improve indoor air quality. This research suggests that ensemble models offer a viable solution for accurate indoor CO₂ prediction, with practical applications in indoor environmental management, especially when coupled with biophilic design elements such as indoor plants.