Optimization of Decoupling Capacitors in Power Delivery Networks Using Machine Learning Algorithms

This research paper presents a comprehensive comparative analysis of six regression algorithms used for frequency prediction in Power Delivery Networks (PDNs). The study evaluates the performance of Decision Tree, Random Forest, Polynomial Regression, Ridge Regression, Lasso Regression, and CatBoost Regression on a dataset containing electrical properties and corresponding frequency values. The models were assessed based on crucial metrics, including Mean Squared Error (MSE), Mean Absolute Error (MAE), and R-squared (R2), to gauge prediction accuracy. Moreover, scatter plots of predicted versus actual frequency values and feature importance plots were utilized to gain valuable insights into the models' behavior. The results indicate significant variations in algorithm performance, with CatBoost Regression demonstrating superior accuracy and emerging as the top-performing method for frequency prediction in PDNs. These findings offer practical guidance for selecting an optimal regression algorithm and hold implications for the implementation of effective power management systems.