Using a machine learning approach to predict the emission characteristics of VOCs from furniture

The emissions of volatile organic compounds (VOCs) from indoor furniture contribute significantly to poor indoor air quality. We have taken a typical machine learning approach using an artificial neural network (ANN), to predict the emission behaviors of VOCs from furniture. The gas-phase VOC concentrations from four kinds of furniture (solid wood furniture, panel furniture, soft leather furniture, soft cloth furniture) were measured in a 1 m³ chamber at different temperatures, relative humidity and ventilation rates. We then used these VOC concentration data as input for training. The trained ANN model could then be used to predict VOC concentrations at other emission time. We selected a back-propagation neural network, with 3 hidden layers, and a learning rate of 0.01. Pearson correlation analysis demonstrates that there is a strong correlation between the input datasets. We used relative deviation (RD) and mean absolute percentage error (MAPE) as the criteria for evaluating the performance of the ANN. For all of the tested VOCs from different types of furniture, the RDs between the predictions and experimental data at 150 h, are less than 15%. The MAPE values of the ANN model are within 10%. These indicate that the trained ANN model has excellent capability in predicting the VOC concentrations from furniture. The main merit of the ANN is that it doesn't need to solve the challenging problem of obtaining the key parameters when using physical models for prediction, and will thus be very useful for indoor source characterization, as well as for exposure assessment.