A new Ensemble based multi-agent system for prediction problems: Case study of modeling coal free swelling index

Abstract In this article, a new ensemble based multi-agent system called “EMAS” is introduced for prediction of problems in data mining. The EMAS is constructed using a four-layer multi-agent system architecture to generate a data mining process based on the coordination of intelligent agents. The EMAS performance is based on data preprocessing and prediction. The first layer is dedicated to clean and normalize data. The second layer is designed for data preprocessing by using intelligent variable ranking to select the most effective agents (select the most important input variables to model an output variable). In the third layer, a negative correlation learning (NCL) algorithm is used to train a neural network ensemble (NNE). Fourth layer is dedicated to do three different subtasks including; knowledge discovery, prediction and data presentation. The ability of the EMAS is evaluated by using a robust coal database (3238 records) for prediction of Free Swelling Index (FSI) as an important problem in coke making industry, and comparing the outcomes with the results of other conventional modeling methods Coal particles have complex structures and EMAS can explore complicated relationships between their structural parameters and select the most important ones for FSI modeling. The results show that the EMAS outperforms all presented modeling methods; therefore, it can be considered as a suitable tool for prediction of problems. Moreover, the results indicated that the EMAS can be further employed as a reliable tool to select important variables, predict complicated problems, model, control, and optimize fuel consumption in iron making plants and other energy facilities.