A Hybrid Machine Learning Approach to Predict and Evaluate Surface Chemistries of Films Deposited via APPJ

Abstract

We developed a hybrid machine learning model, integrating Artificial Neural Network (ANN), Random Forest (RF) and AdaBoost (AB), to predict and evaluate the plasma polymerization process of TEMPO monomer, specifically for Nitric Oxide films. This model is specifically designed to adeptly navigate the intricate landscape of the plasma polymerization process. Through genetic algorithm optimization, we have fine-tuned our hybrid model's algorithm weights, achieving results that closely match experimental data. TEMPO-Helium flow ratio is identified as the most critical parameter for the surface N percentage, with a relative importance of 41%. Frequency has the greatest influence on the N-O percentage, with a relative importance of 30%. The intertwined influence of different polymerization parameters on the film's surface chemistry has been detailed.