This research aims to analyze the effectiveness ofthe Class-Rebalancing Self-Training (CReST) method in semisupervisedlearning (SSL) on class-imbalanced data. The studyuses the CIFAR 10 long-tailed dataset to test the performance ofSSL with CReST using Python programming language on theGoogle Colab platform. The results showed that CReSTeffectively reduces pseudo-labels in the majority class andincreases recall in the minority class, with the best performanceachieved at Generation 16. However, there was a decrease inAverage Accuracy Recall per Class after Generation 16. Thestudy suggests addressing the over-sampling issue and exploringthe application of the CReST framework in other areas ofmachine learning and AI.Kata kunci— CReST, Semi-Supervised Learning, imbalancedata, pseudo label, Semi-Supervised Learning Generation