Predicting Intersystem Crossing Rate Constants of Alkoxy-Radical Pairs with Structure-Based Descriptors and Machine Learning

This repository contains datasets and machine learning code for predicting intersystem crossing (ISC) rate constants in radical pair systems. The data includes geometries, spin-orbit couplings, excitation energies, and ISC rates for 98,082 conformations of ten different alkoxy radical dimers. Three ML models—Random Forest, CatBoost, and a feed-forward neural network—were trained using geometrical descriptors as inputs. Scripts for hyperparameter optimization, feature selection, and evaluation are also provided.