A New Approach for Bias–Variance Analysis Using Regularized Linear Regression

Reddy, M. Rajasekhar; Kumar, B. Nithish; Rao, N. Madhusudana; Karthikeyan, B.

doi:10.1007/978-981-15-0339-9_4

Cited by 5 publications

(1 citation statement)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The outputs are n e and T e , which minimize the object function. An optimum value of µ is determined via bias-variance analysis [25][26][27]. The analysis helps us understand the trade-off between the new model's ability to fit the measured line intensity well (low bias) and its ability to generalize accurate electron density, electron temperature, ground-state density, and escape factors (low variance).…”

Section: Bias-variance Analysismentioning

confidence: 99%

Improved Line Intensity Analysis of Neutral Helium by Incorporating the Reabsorption Processes in a Helium Collisional-Radiative Model

Lin

Goto

Akatsuka

2023

Atoms

View full text Add to dashboard Cite

In this study, eight emission lines in the visible wavelength range of neutral helium were used to diagnose the electron density and temperature of the Large Helical Device (LHD) helium plasma instead of the conventional three-line method. The collisional-radiative (CR) model for low-pressure helium plasma was revised to include the optical escape factors for spontaneous transition from the n1P states to the ground state so that the influence of the absorption effect under optically thick conditions could be considered. The developed algorithm was based on fitting the number densities of eight excited states obtained using optical emission spectroscopy (OES). The electron density, electron temperature, ground-state density, and optical escape factors were selected as the fitting parameters. The objective function was set as the summation of the residual errors between the number densities measured in the experiment and those calculated using the revised model. A regularization term was introduced for the optical escape factor and optimized through bias and variance analyses. The results show that the agreement between the number density calculated by the algorithm and its counterpart measured in the experiment was generally improved compared to the method using three lines.

show abstract

Section: Bias-variance Analysismentioning

confidence: 99%