TY - GEN
T1 - Predicting The Melting Temperature Of Polymers Using Machine Learning
AU - Fatriansyah, Jaka Fajar
AU - Mahardi, Hanif Lanang
AU - Rizky, M. Ali Yafi
AU - Surya, Robertus Darwin
AU - Federico, Andreas
AU - Surip, Siti Norasmah
N1 - Publisher Copyright:
© 2024 IEEE.
PY - 2024
Y1 - 2024
N2 - One crucial factor in polymer material fabrication is its melting temperature, which is key to determining the manufacturing process. However, difficulties have been encountered in finding correlations between the melting temperature and influencing factors. Therefore, machine learning is employed to discern the relationship between polymer structure and its melting temperature. The models used include KNN, SVR, and XGB, as well as deep learning with ANN and NLP. The ability of SMILES and NLP descriptors in describing polymer structures is also examined. The input length of the models varied between 50 and 200. All models produce the most accurate predictions with an input length of 200. KNN and SVR generate relatively stable but consistently low R-squared values, with MAE values above 40. ANN and NLP exhibit high R-squared values in some trials but with low stability and relatively high MAE values. XGBoost consistently produces higher and more stable R2 values. The best-performing XGBoost model is validated using SMILES strings outside the dataset, demonstrating its predictive ability with an MAE of 24.09, deemed sufficient for predicting melting temperatures.
AB - One crucial factor in polymer material fabrication is its melting temperature, which is key to determining the manufacturing process. However, difficulties have been encountered in finding correlations between the melting temperature and influencing factors. Therefore, machine learning is employed to discern the relationship between polymer structure and its melting temperature. The models used include KNN, SVR, and XGB, as well as deep learning with ANN and NLP. The ability of SMILES and NLP descriptors in describing polymer structures is also examined. The input length of the models varied between 50 and 200. All models produce the most accurate predictions with an input length of 200. KNN and SVR generate relatively stable but consistently low R-squared values, with MAE values above 40. ANN and NLP exhibit high R-squared values in some trials but with low stability and relatively high MAE values. XGBoost consistently produces higher and more stable R2 values. The best-performing XGBoost model is validated using SMILES strings outside the dataset, demonstrating its predictive ability with an MAE of 24.09, deemed sufficient for predicting melting temperatures.
KW - ANN
KW - KNN
KW - Machine learning
KW - MAE
KW - Melting Temperature
KW - NLP
KW - Polymer
KW - R-squared score
KW - SVR
KW - XGBoost
UR - http://www.scopus.com/inward/record.url?scp=85213319173&partnerID=8YFLogxK
U2 - 10.1109/iSemantic63362.2024.10762597
DO - 10.1109/iSemantic63362.2024.10762597
M3 - Conference contribution
AN - SCOPUS:85213319173
T3 - Proceedings - 2024 International of Seminar on Application for Technology of Information and Communication: Smart And Emerging Technology for a Better Life, iSemantic 2024
SP - 307
EP - 311
BT - Proceedings - 2024 International of Seminar on Application for Technology of Information and Communication
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2024 International of Seminar on Application for Technology of Information and Communication, iSemantic 2024
Y2 - 21 September 2024 through 22 September 2024
ER -
