Application of Machine Learning Methods in Predicting Chemical Solubility

Abstract

Solubility is an important property in chemistry, especially in drug development, as it affects how well a drug can work in the human body. Traditionally, scientists predicted solubility by studying the structure of molecules, but these methods are not always accurate. In recent years, machine learning has become popular in many fields, and this study explores how it can help improve solubility prediction. A dataset from Kaggle with four features including MolLogP, MolWt, NumRotatableBonds, and AromaticProportion was used. The study tested three models: Random Forest, Decision Tree, and Linear Regression. The data was split into training and test sets. Among the models, Random Forest performed best, with an accuracy of 0.90 and AUC of 0.89. It also showed that MolLogP was the most important factor affecting solubility. Visual tools like heatmaps and boxplots helped explain the results clearly. This research shows that machine learning is a useful tool in chemistry and can support better drug design in the future.