AI-Assisted Design of Power Electronic Circuits

In order to assess the performance of the proposed AI assisted approach incorporated with data normalization, Recursive Feature Elimination (RFE), Random Forest Classifier, compared against other conventional machine learning approaches, the approach was evaluated. The Random Forest model achieves the following accuracy, precision, recall and F1-score: 95.2%, 94.6%, 93.8%, and 94.2% respectively. As compared to that, a Support Vector Machine (SVM) model using the same dataset and preprocessing obtained an accuracy of 89.5% and a Decision Tree classifier was 90.1%. Also, an accuracy of 88.7% with low precision and recall was reported in a K Nearest Neighbors (KNN) model. RFE use significantly improved the model performance in all the methods, the largest gain was observed in the Random Forest where it has shown that it handles non linear interactions and high dimensional data well. Additionally, feature importance analysis in Random Forest gave explainability, however, not in case of SVM and KNN. In both predictions and interpretability, it clearly shows that integration of RFE with Random forest is stronger compared with traditional models. The result is validated, which portrays a great effectiveness of the proposed AI based workflow, in optimizing the circuit design parameters, and accompanying engineers to make wise design choices with high confidence.