نوع مقاله : مقاله پژوهشی
نویسندگان
دانشکده مهندسی معدن و متالورژی، دانشگاه صنعتی امیرکبیر تهران
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Summary
Importance of detecting anomalous values from background is undeniable; so many different methods have been developed. On the other hand, prediction is a powerful tool in the process of each task’s planning. Using data mining methods for finding patterns between data can answer this requirement. Due to the necessity of high processing accuracy, the performance of a separation method has been evaluated. This performance is combined with three data mining methods. Finally we introduced the best combination (Mahalanobis Distance and Decision Tree) with the most accurate predictions. In order to reduce error and risk to save costs, time, energy and access to the more valuable predictions, in following paper we have studied Mahalanobis Distances to separate anomalous values and combined the results with three methods: K–Nearest Neighbor (KNN), Naïve Bayes Classifier and Decision Tree (DT) then achieved to the best combination with the least error rate.
Introduction
Mahalanobis distances method was used to assess prospective areas of Parkam district based on the two variables (Cu and Mo grade) and anomalous values have been defined based on the grades of copper and molybdenum. Then the three mentioned algorithms were trained by 4 parameter data grades of copper and molybdenum, coordinates of each samples (X and Y) and the results of our separation method as well, finally predictive equations were achieved in order to inform about other random samples. The best combination could be useful to predict with high accuracy in each plan.
Methodology and Approaches
Mahalanobis distances method is an effective multivariate method on separation of anomalous values from background. Various data mining methods have been developed to classify data. Three most important and common methods are K–Nearest Neighbor, Naïve Bayes Classifier and Decision Tree; They can be used to find features that can distinguish different classes from each other.
Results and Conclusions
After separation of anomalous values by applying Mahalanobis distances, combined models have been produced. Then actual data have been seen as the test ones to evaluate the accuracy of predictions. At last, based on the resubstitution rate, that is 0.0053, for designed system via Decision Tree technique and anticipating only 2 out of 377 numbers of samples as the background samples instead of anomalous ones, this method was recognized as the more pragmatic approach than KNN and Naïve Bayes approaches producing 0.0239 and 0.061 error rate and predicting 9 and 23 numbers of anomalous values as the background samples respectively. According to the much more acceptable error rate for designed network by combination of Mahalanobis Distances and Decision Tree methods, we can introduce that as a much more reliable and useful method in order to achieve the most accurate predictions to the decision makers in the industry.
کلیدواژهها [English]
)