Predict Efficiency of Organic Solar Cell with Low Generalization Error Based on Molecular Property and Device Fabrication

Organic solar cells (OSCs) have been a very active research field in recent years. There are two main optimization strategies which are novel donor or acceptor and device fabrication. Due to the huge number of influencing factors and their complicated internal interaction mechanism, it's almost impossible to build a complete theory to describe and analyze device power conversion efficiency (PCE). However, machine learning may be a feasible answer. In this research, molecular properties and device fabrication are combined to build dataset. To decrease generalization error, models are developed by random forest, support vector machine and multiple perceptron. Random forest shows the best performance and is determined to the final algorithm. After further optimization, the test set R-2 average of 100 different random state converges on 0.9012 and the quantitative results of feature importance are given. The dataset plays a critical role in the performance of machine learning model. The results indicate the feasibility of applying results given by machine learning models as references for experiments and analysis.