Classification and prediction of the mesophases of block copolymers

Self-organization and the resultant mesoscopic structures in block copolymer systems are of great advantage to applications such as nanolithography. Nevertheless, the whole phase behaviors still have not understood well even in ABC triblock copolymers. We study on classification problem on the classical phase diagram of AB diblock copolymers by means of supervised machine learning. We compare three model: kernel support vector machine (SVM), random forest, and k-nearest neighbor method. The prediction accuracy of kernel SVM is 94.5%, which is higher than the other two methods. This indicates that kernel SVM can be a candidate of classification model that is applicable to more complex architectures, such as linear/star ABC triblock copolymers.