Predicting the scalar coupling constants between atom pairs in a molecule through machine learning applications
Sean Kim
Fort Lee High School, USA
: J Comput Eng Inf Technol
Abstract
In this paper, we used data on various molecular characteristics from the CHAMPS Kaggle competition (CHAMPS, 2019) to build a prediction model based on various machine learning approaches. We built models on the training set (N = 4,659,076 observations) and then used the best performing one to obtain and evaluate predictions on the testing set (N = 2,505,190 observations). We evaluated the performance of three models – linear regression, XGBoost, and Neural Net – on three metrics: R-squared, MAE, and RMSE. The XGBoost model resulted in a superior fit over Neural Nets and linear regression, with RMSE as lower as 2.75 on the test dataset. This result suggests that XGBoost is a viable approach for predicting the scalar coupling constant. Keywords: Scalar coupling constants, atom pairs, machine learning, data science, prediction model Recent Publications 1. Kim, S. (2022) “A Token-Based Voting System for Rating News Sources” presented and published at The Artificial Intelligence & Robotics Conference in Osaka, Japan.
Biography
Sean is a senior at Fort Lee High School in New Jersey, USA. He is passionate about computer science and hopes to conduct more research in the field.
