A Machine Learning Approach for Polymer Classification Based on the Thermal Response under Data ScarcityTested on PMMA

An important application of machine learning techniquesis theintelligent nondestructive testing of polymers. However, data scarcityand class imbalance (for real applications) shape some of the involvedchallenges. In order to tackle these challenges, an intelligent screeningframework for poly(methyl methacrylate) (PMMA) samples is studiedhere. An efficient thermal and experimental test is designed and coupledwith several machine learning tools for quick classification withfour features. Furthermore, a set of sampling techniques are employed/compared;these techniques are Random oversampling (ROS), synthetic minorityoversampling technique (SMOTE), and adaptive synthetic sampling (ADASYN).Then, Linear Discriminant Analysis, K-Nearest Neighbor, Naive Bayes,decision tree, random forest, pattern recognition network, supportvector machine, and ensemble learning are employed for classification.For assessing these algorithms, their performances are evaluated usinga collection of metrics (i.e., Geometric-mean, F1 score, Matthewscorrelation coefficient, accuracy, true positive rate, true negativerate, positive predictive value, and negative predictive value). Amongothers, the average G-mean measures, which is a paramount measurefor assessing the imbalance data, are increased from 75.97% (originaldata) to 94.24% (ROS), 93.49% (SMOTE) and 91.27% (ADASYN). That isa clear proof of successful oversampling. The final results show thatROS oversampling coupled with Ensemble classification methods cansignificantly improve all performance metrics for PMMA classification.

Automated summary

An intelligent screening framework for PMMA samples is developed using machine learning tools and various sampling techniques. The performance of different classification algorithms is evaluated using multiple metrics. The results show that ROS oversampling coupled with Ensemble classification methods significantly improves the performance metrics for PMMA classification.