Mechanical properties such as tensile strength, ductility, and tensile modulus are essential criteria in polymer matrix composites (PMC) design and are determined through the stress-strain curve obtained from the tensile test. Material designers can examine the stress-strain curve trends based on the combination and composition, but it is difficult to predict using numerical analysis software due to the complex correlation based on chemical properties. To address these limitations in PMC design, this study uses feature engineering methods such as principal component analysis (PCA) and recursive feature elimination with cross validation (RFECV) to find the minimal and optimal set of features necessary for predicting the tensile behavior of PMC. The Long Short-Term Memory (LSTM) and feedforward neural network (FNN) models are trained using the optimal feature set and 1,270 PMC’s tensile test data to predict the tensile stress-strain curve. The predictive model developed in this study provides stress-strain curves of tensile tests, including tensile failure of PMC, which can be challenging due to the high nonlinearity of PMC. Material designers can reduce the time and labor costs of PMC design through this tensile behavior prediction model that has an accuracy of R2 =92% and requires fewer features. In addition, the model can be used as a high-throughput screening model for PMC inverse design systems.