Prediction of Air Quality Index by Extreme Learning Machines

In this study, it is aimed to predict the Air Quality Index (AQI) by the Extreme Learning Machines (ELM) algorithm. For this purpose, six parameters have been selected which can affect the AQI. These are temperature, humidity, pressure, wind speed, PM10 and SO
₂
respectively. First of all, “Forecast Sheet” application that presented in the Excel environment and the correlation analysis were used to determine the relationship status between these six parameters and AQI. Thus, the linear equations between each parameter and AQI were determined. Also, R values were determined for relation rate. Then, a training data set was created. 120 values belong to March, June, September and December of 2017, which includes six parameters, was determined as training data. AQI values of these six parameters were obtained by “AQI Calculator” application. The obtained AQI values were classified mathematically from 1 to 6 considering the AQI limits determined by National Air Quality Monitoring Network of Ministry of Environment and Urbanization. The first 12 day parameters of January, March, September and December of 2018 year are determined as test data. Then, ELM used 108 of the 120 generated data as training data and the remaining 12 of the data as test data. 10-fold cross validation method was used for the reliability of the result. Sine, sigmoid and hard limit activation codes were used for classification. 6 different neurons were studied. As a result of cross-validation, the activation code and the number of neurons that achieved the highest success were applied to the actual test data. Finally, the classification values predicted by ELM and the mathematical classification values were compared. Thus, the success of the ELM was measured.