A New Hybrid Data Mining Technique to Forecast the Greenhouse Gases Emissions

The expansion of industrial activities and the unnecessary growth of cities have increased the concentration of greenhouse gases, including carbon dioxide in the atmosphere. Mostly, CO2 emissions are caused by the consumption of different forms of energy and the combustion of all types of fuels, especially fossil fuels. The development of data mining techniques that lead to accurate prediction of CO2 emissions is very useful in deciding the preventive measures and appropriate policies in this area. Most studies in this field are limited to models that do not compare different techniques and features and only examine the effect of economic factors and fossil fuel consumption on CO2 emissions. The aim of this study is to identify a combination of significant features as well as to select the best technique to predict CO2 emissions. For this purpose, a huge dataset containing various features was obtained from the IEA database. A new hybrid method for predicting CO2 emissions was developed, and then results were compared with proposed data mining techniques including ANN, KNN, GLE, Linear-AS, and Regression. Also, a combination of significant features and the best techniques for predicting CO2 emissions were identified. The results of the proposed method show that by clustering the database and then implementing prediction techniques, the error could be substantially reduced. Also, in order to predict future observations, first, they are placed in appropriate clusters with the Discriminant Analysis technique and then they are predicted with the appropriate forecasting technique. It was found that the proposed hybrid technique, which is a combination of K-Means, Linear-AS and Discriminant Analysis, is most accurate in this case.