Predicting the complex stress-strain curves of polymeric solids by classification-embedded dual neural network

•A dual neural network architecture is proposed to predict different types of stress–strain curves.•This model is featured by the state-of-the-art simplicity of 300 neurons in total.•Such simplicity allows the model to utilize a very small dataset of 27 samplings from a 4D space.•This model is easily interpreted to gain physics insights into the mechanical behaviors.