College
College of Engineering and Polymer Science
Date of Last Revision
2026-05-07 06:09:33
Major
Chemical Engineering
Honors Course
CHEE: 497
Number of Credits
3
Degree Name
Bachelor of Science
Date of Expected Graduation
Spring 2026
Abstract
The goal of this project was to utilize the Signature molecular descriptor to construct a predictive model for computing the normal boiling point using a 4856-compound dataset. The final model intends to be applied to screening compounds whose experimental normal boiling point values are not readily available. Compounds were deconstructed into their height-1 atomic Signatures, resulting in 225 unique descriptors for the dataset, which were used as independent variables for creating the model. Machine learning (ML) techniques were employed to evaluate the applicability and accuracy of different ML models for predicting normal boiling point using the Regression Learner App in MATLAB for training and testing. 26 total ML models were evaluated with multiple linear regression (MLR), support vector machines (SVM), and Gaussian process regression (GPR) models performing with the greatest predictive accuracy. Testing R2 values for these models were 0.84, 0.86, and 0.88, respectively. Although the more complex SVM and GPR models exhibited greater predictive ability, MLR was selected as the optimal choice for the final model due to its simplicity, ease of interpretability and implementation in accordance with its intended use.
Research Sponsor
Dr. Donald Visco
First Reader
Dr. Fardin Khabaz
Second Reader
Patrick Cuddihy
Honors Faculty Advisor
Dr. Bi-min Zhang Newby
Proprietary and/or Confidential Information
No
Community Engaged Scholarship
No
Recommended Citation
Karabin, Mackenzie, "Development of a Signature Model for Predicting Normal Boiling Point Using Machine Learning" (2026). Williams Honors College, Honors Research Projects. 2197.
https://ideaexchange.uakron.edu/honors_research_projects/2197