IndoAirSense: A framework for indoor air quality estimation and forecasting

Praveen Kumar Sharma,Ananya Mondal,Shivam Jaiswal,Mousumi Saha,Subrata Nandi,Tanmay De,Sujoy Saha

DOI: https://doi.org/10.1016/j.apr.2020.07.027

IF: 4.831

2021-01-01

Atmospheric Pollution Research

Abstract:<p>Studies reveal that poor indoor air quality has resulted to a drop in academic performance and affected the comfort and concentration level of students consequently impacting their learning ability during the lecture. Most of the studies on indoor air quality in university settings largely focus merely on the measurement of parameters like temperature, humidity, <span class="math"><math>CO2</math></span> and <span class="math"><math>PM2.5</math></span>, etc., primarily dealing with perception-based study. Moreover, existing works consider installation of sensors in each and every lecture hall which incurs huge cost, hence may not be feasible for many Universities in low and medium economy countries. This paper proposes <em>IndoAirSense</em>, a low-cost, effective framework for indoor air quality estimation and forecasting in the context of university classrooms. In contrast to the existing approaches, assuming similar ambient and structural features of classrooms, this work considers deploying suitably-calibrated low-cost sensors, in only a few selected classrooms of the building. Using the sensed parameters, we apply Multi-Layer Perceptron (MLP) and eXtream Gradient Boosting Regression (XGBR) for real-time estimation of the air quality in other rooms without sensors. Finally, prediction of the air quality of the rooms are performed using LSTM-wf a modified Long Short Term Memory (LSTM) machine learning model and we achieved accuracy of nearly 95% and 96% during estimation and forecasting, respectively. An experiment on 10 classrooms involving 28 instructors and 60 students for a period of 256 h (spanning 32 days) has been performed using the developed application. The system App recorded user feedback on the perceived air quality revealing nearly 76% three stars (high correlation with system results) and 15% two stars (medium correlation) reviews, which validates the effectiveness of our framework.</p>

environmental sciences