Plasma Density Prediction for Helicon Negative Hydrogen Plasma Source Using Decision Tree and Random Forest Algorithm

Abstract

Plasma density of an ion source is an important parameter to be measured for the characterization of an ion source. Measurement of plasma density depends on several input parameters such as input power, gas pressure, magnetic, plasma potentials, and electron temperature. In the absence of a direct mathematical relationship between the input parameters and plasma density, measuring plasma density becomes a time-consuming task and has been done through invasive experimental methods. This paper utilizes decision tree and random forest algorithms for developing a predictive model of a negative hydrogen helicon plasma source to estimate the plasma density from the RF input power, magnetic field (B), and gas pressure. For this purpose, a dataset having 162 samples have been applied for training, testing and validation of the proposed models. A comparative analysis of proposed models has been done through by comparing of root mean square error and the coefficient of determination (R2) between the measured and the predicted values of plasma density. The obtained results show that the random forests model has shown better results and able to predict plasma values closer to the actual experimental results. Also, random forests algorithm can interpret the complex non-linear relationships between the influential input parameters and the output.