Predictive modeling and optimization of pin electrode based cold plasma using machine learning approach

Cold atmospheric pressure plasma (CAP) is a technology with immense potential in various technological and bio-medical domains. The present paper proposes a statistical and machine learning-based modeling and optimization methodology for a novel pin electrode based atmospheric pressure cold plasma jet (APCPJ), with focus on its operation in the glow discharge region, because of its relevance in biomedical applications. A feedforward backpropagation artificial neural network (ANN) model is developed in capturing the relationship between the input parameters of supply voltage (SV) and frequency (SV) with the performance parameters, power consumption and jet lengths (with and without sleeve). The robustness of the developed ANN model is demonstrated by predicting the performance parameters of the CAP within and beyond the experimental range. The composite desirability approach is utilized to obtain the optimized settings of SV and SF for simultaneous maximization and minimization of the jet lengths (with and without sleeve), and power consumption, respectively. Finally, three machine learning models of logistic regression, viz., K-nearest neighbor (KNN), discriminant analysis (DA) and ANN classifier (ANNC) are implemented to classify the discharge regions of the generated plasma whose accuracy is depicted using the confusion matrix and the receiver operating characteristic curves.