Assessing the Efficacy of Symbolic Regression for Scientific Data Classification and Regression

Symbolic regression (SR) is a subdomain of evolutionary computing which aims to generate mathematical equations which best fit a given dataset. This is done with the help of genetic algorithms. Symbolic Regression is especially useful in cases where features in a dataset are known to have distinct mathematical relations to each other and its output. Therefore, this paper aims to compare and provide insight on the results of using SR on two datasets with two python libraries. These results are further compared with other ML models to assess their performance in classification and regression tasks. The detailed methodology and results are discussed further test metrics and obtained equations. The results show that symbolic regression can perform very well on certain datasets whose target outputs have a mathematical connection to the input features. SR is found to work better than other ML models in the classification and regression tasks.