Document Type : Original Article

Authors

• Mohammadsadegh Behrouz ¹
• Mohammad Ali Afshar Kazemi ²
• Adel Azar ³
• Ezatollah Asgharizade ⁴

¹ Ph.D. student of Industrial Management, Faculty of Management and Economics, Branch of Science and Research, Islamic Azad University, Tehran, Iran.

² Associate Professor, Faculty of Management, Branch of central Tehran, Islamic Azad University, Tehran, Iran.

³ Professor, Management and Accounting Faculty, Tarbiat Modares University, Tehran, Iran.

⁴ Associate professor, Faculty of management, Tehran University, Tehran, Iran.

Abstract

Predictive approaches in the field of risk management plays an important role in detecting events, controlling risks and reducing maintenance and repair costs. The purpose of this research is to provide a model for predicting critical and prioritized risks based on data mining algorithms. The historical data of the implementation of maintenance programs and risk assessment were used and the data mining method was planned based on the CRISP methodology. Data modeling has been done in two parts: "descriptive" and "predictive" data mining and the use of "clustering" and "classification" algorithms. After data pre-processing and preparation, the "silhouette index" is considered as the basis for clustering and the K-Means, Kohnen, Two Step algorithm is used; the best value is based on the K-Means algorithm. Silhouette is equal to 0.6446 with the number of clusters 5, and the main characteristics have been determined for the classification and prediction of risks. Next, Neural Network Algorithms, C.5 tree, Nearest Neighbor and Support Vector have been used for classification. These techniques are widely used in machine learning and recognizing data classification patterns and their integration increases the amount of data learning. The results showed learning in 97.56% of the agreed data and the accuracy and validity of the combined model for data classification was estimated at 92.86%. Based on the results, 13 critical risks have been identified; "release of polluting gases and chemicals" and "lack of training and justification of contractors regarding the pipeline" have the highest and lowest priority, respectively.

Keywords

• Risk Assessment
• Modeling
• Data Mining
• Gas pipeline

Main Subjects

• Data Mining