Explaining of prediction accuracy on phase selection of amorphous alloys and high entropy alloys using support vector machines in machine learning

To explain the different prediction accuracies of a single characteristic parameter in amorphous alloy (AM), solid solution alloy (SS) and high entropy alloy containing intermetallic compound (IM), three methods were proposed. The first was that the simple division in the whole value range of the characteristic parameter can qualitatively explain the high or low prediction accuracies of characteristic parameters in three type of AM, SS and IM alloys. To consider the mutual interference to the decision boundary in the whole eigenvalue range, the second was that the histogram of the probability density distribution of eigenvalues was used to qualitatively explain the prediction accuracies of characteristic parameters in different regions. The third was that the Gaussian fitting curves of the histogram of probability density distribution for characteristic parameters was considered. The prediction accuracies of the characteristic parameters in different alloys were explained by using the method of normalized area or the sum of the normalized areas of the two characteristic parameters. The analysis results were basically consistent with the results of model learning. To explain the prediction accuracies of two or three parameter combinations, an expression was defined to simply express the interference ability of the single characteristic parameter to the decision boundary division. The comparison of the prediction accuracies of two or three parameter combinations was analyzed by the expression of the decision-boundary interference ability of a single characteristic parameter. Three parameter combinations had the highest average prediction accuracy, however, the average prediction accuracies of four parameter combinations had been over fitted for AM, SS and IM alloys. This work provided a new interpretation method for the prediction accuracy of phase composition of AM, SS and IM alloys based on support vector machines in machine learning.

Automated summary

Three methods were proposed to explain the different prediction accuracies of a single characteristic parameter in amorphous alloy (AM), solid solution alloy (SS), and high entropy alloy containing intermetallic compound (IM). The first method used simple division to qualitatively explain the high or low prediction accuracies. The second method used histograms of eigenvalues' probability density distribution to explain the accuracies in different regions. The third method involved Gaussian fitting curves of the histograms. The analysis results were consistent with the model learning results.