Benzene Air Pollution Monitoring Model using ANN and SVM

Air Pollution considered to be a very significant issue concerning the people and the environment. Therefore, many countries develop a monitoring system in order to reduce the harm caused by the pollution of air. Monitoring helps us to assess the presence of harmful pollutants in the air, such as CO, NOx, NO 2 and C6H6. To have a better-quality air means having better quality of life. In this paper, we train a dataset that measures the hourly average of five metal oxide chemical sensors implanted into an Air Quality Chemical Multi sensor Device along with temperature, humidity and artificial humidity. We will estimate the concentration of benzene according to its correlation with CO, we use two models, namely Artificial Neural Networks (ANN) and Support Vector Machine (SVM) running on MATLAB. Using ANN, we divide the dataset into 10 days, 20 days, three months and 30 weeks to calculate the Mean Relative Error and Mean Absolute Error. We found that the true concentration of the benzene was close to its estimation with Mean Relative Error less than 20% which indicates a good result compared with the previous works, while when using SVR the Mean Relative Error was almost 30% which gives a worse performance. However, SVR might be a better model as the Errors decreased regularly when increasing the dataset length, which makes a better prediction model unlike ANN where the errors were unstable when increasing the dataset length.