Application of MARS for the Construction of Nonparametric Models

2000 Mathematics Subject Classification: 62G08, 62P30.

This paper presents the main features of the relatively new statistical technique called Multivariate Adaptive Regression Splines (MARS) and the corresponding software product. The MARS method is designed for statistical analysis of data, when standard parametric modeling by multiple regression or logistic regression methods is not applicable. A case study from the area of laser technology, especially for modeling of UV Cu+ Ne-CuBr laser is performed. The obtained results are in a good agreement with practical issues. It is shown that the constructed nonparametric MARS models can be applied in estimation and prediction of current and future experiments in order to improve the output laser power.

This paper is partially supported by projects VU-MI-205, NSF of the Bulgarian Ministry of Education, Youth and Science and RS09–FMI–013, ISM-4 of NPD, Plovdiv University “Paisii Hilendarski”.