In the present study, the structure-property relationships (SPR) are investigated using finite element analysis (FEA) and machine learning (ML) methods. The objective of the research was to create an ML model capable of predicting the temperature-dependent stress-strain response of two-phase composites, as well as the amount of heat generated during plastic deformation considering different Taylor-Quinney coefficients (TQCs). The ML model aims to achieve industrially acceptable prediction accuracy and significantly better computational efficiency compared to FEA. Microstructure generation and FEA were used to create the datasets for the ML model, and both processes were automated using the Python programming language. FEA was performed for each representative volume element (RVE) to obtain the stress-strain response as well as the temperature increase-strain curve. This means that two separate datasets and two separate ML models were created: one for the stress-strain responses and one for the temperature increasestrain responses. Both ML models were developed in the form of complex 3D-convolutional neural networks (CNNs), taking the microstructural configuration as input. Other input parameters included the TQC value and temperature for the first and second ML models, respectively. The outputs of the CNNs were the temperature increase-strain curves and the stress-strain curves at specific temperatures for the first and second model, respectively. Compared to FEA, both the model for predicting temperature-dependent stress-strain curves and the model for predicting plastic deformation-induced heat generation demonstrated very good prediction accuracy suitable for industrial applications. Additionally, both models exhibited significantly faster prediction capabilities and better computational efficiency than FEA.