It is widely known that small defects play a significant role in metal fatigue. For instance, even if the size of the defect is too small to be seen by the naked eyes, it has a significant negative impact on fatigue limit. Today, AI technology has progressed to the point of various practical applications, and we see more and more AI-based services in our daily lives. In practice, various machine learning methods are available to support the development of AI. Given this circumstance, it is expected that the machine learning can be applied to solve the problems related to metal fatigue.

In our previous study, by taking advantage of essence of the area-parameter model, we showed that Ridge regression (L2 regularization) is applicable to estimate the fatigue limits with small defects. Moreover, L2 regularization is useful to enhance the regularity of the matrix equation and to control the accuracy of estimation. As an extension of the previous study, in this study, by using the multivariate polynomials in a systematic manner, Lasso regression (L1 regularization) is applied to estimate the fatigue limit. Further, by performing the cross-validation, the influence of a hyperparameter and the generalization performance are examined.