http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
Automatic Classification Models and Algorithms Based on the Minimum Sum-of-Squared Errors Model
Guzel Shkaberina,Lev Kazakovtsev 대한산업공학회 2020 Industrial Engineeering & Management Systems Vol.19 No.4
We propose new models and algorithms for automatic classification of objects (clustering) based on the minimum sum-of-squared errors clustering (MSSC) model. Our approach was aimed at improving the accuracy and stability of the result in solving practical problems, such as identifying homogeneous batches of industrial products. We examined the application of the MSSC model and k-means algorithm with various distance measures: Euclidean, Manhattan, Mahalanobis for the problem of automatic classification of objects in a multi-dimensional space of measured parameters (features). For such problems, we present a new model (Mahalanobis Minimum Sum-of-Squared Error Clustering, MMSSC) for solving problems of automatic classification based on the MSSC model with Mahalanobis distance. In addition, we present a new algorithm for automatic classification of objects based on the MMSSC optimization model with the Mahalanobis distance measure and the weighted average covariance matrix calculated from the training sample (pre-labeled data). This algorithm allows us to reduce the number of errors (increasing the Rand index) when identifying homogeneous production batches based on the results of quality control tests. A new approach in the development of evolutionary algorithms for the MSSC problem is presented using a greedy agglomerative heuristic procedure contained in several genetic operators. The use of this approach enables a statistically significant increase in the accuracy of the result (the achieved value of the objective function within the chosen MMSSC mathematical model), as well as its stability, in a fixed time, in comparison with the known algorithms. Thus, in this work, an increase in the accuracy of solving the problem of automatic classification is achieved both by increasing the adequacy of the model (according to the Rand index) and by improving the algorithm that allows us to achieve the best objective function values of within the framework of the chosen model.