PREDICTION OF BOILING POINTS OF ORGANIC COMPOUNDS USING MACHINE LEARNING ALGORITHMS

Accurate knowledge and understanding of the thermophysical properties of organic compounds are essential to optimize their performance in industrial applications. In this context, the use of machine learning methodologies represents a promising alternative to expensive experimental tests and complex mathematical models traditionally employed in the determination of thermodynamic properties. This study aimed to evaluate the application of machine learning algorithms to estimate the boiling point of simple organic compounds. A database was developed containing ten substances, including hydrocarbons and oxygenated compounds, whose properties were obtained from bibliographic and public data sources. The predictor variables considered were molar mass, number of carbon atoms (N_C), number of oxygen atoms (N_O), and polarity (binary variable). Analyses were conducted in a Python 3.12 environment using the pandas and scikit-learn libraries. Two supervised regression algorithms were employed: linear regression and decision tree. The dataset was divided into 80% for training and 20% for testing, and model performance was evaluated using the coefficient of determination (R²) and mean absolute error (MAE). The results indicated that the linear regression model presented poor performance (R² = -6.92; MAE = 206.4 °C), demonstrating the absence of a simple linear relationship between the selected molecular variables and the boiling point. In contrast, despite the very limited dataset, the decision tree model showed significantly better results (R² = 0.72; MAE = 54.5 °C), revealing a higher ability to capture nonlinear relationships and interactions among the variables. These findings reinforce that nonlinear learning models are more suitable for predicting complex physicochemical properties, even when applied to small datasets. It is concluded that the application of machine learning algorithms constitutes an efficient, accessible, and didactic approach for predicting thermodynamic properties, with potential use in educational and research activities in chemical engineering, food engineering, and data science.