Despite the usage of machine learning accelerating the properties prediction of polymer materials, obtaining a large number of samples to achieve accurate and fast predictions remains a challenge because of the complex and lengthy experimental process. In this work, an advanced prediction model for the small sample analysis is presented by an ensemble learning algorithm called extreme gradient boosting (XGBoost) based on nearest neighbor interpolation (NNI) and synthetic minority oversampling technique (SMOTE). Different from directly using small sample prediction algorithms, a brand‐new idea based on feature engineering is proposed. NNI algorithm is used to interpolate the original data set to solve the data insufficiency. SMOTE algorithm is used to solve the data imbalance problem by increasing minority samples. The expanded data set is used to build an XGBoost prediction model. A model for predicting Akron abrasion of rubber through mechanical properties is established via the proposed method. The original data set is expanded to 710 samples through two different interpolations. Experimental results show that better prediction accuracy and generalization ability are obtained than traditional algorithms. The Akron abrasion is found to be the most related to the elongation at break of polymer materials.