The reliable design of fiber-reinforced epoxy composites is still challenging because of their complex heterogeneous nature. The limitations of existing predictive models are the joint estimation of tensile and flexural properties for both pure and hybrid constituents. Traditional characterization of these properties is dependent on costly time-consuming destructive testing and, at the same time, simulated predictive models are prone to suffering from oversimplified assumptions. To fill this gap, machine learning (ML) models have been implemented in this study with experimental data of 54 laminates with different fiber constituents (pure, bi- and tri-hybrids), stacking sequences and ply count. ML models included baseline, ensemble and neural networks, which were trained, validated and tested where design and testing parameters were input features for the prediction. Experimentally, it was observed that Kevlar -cross 4 Ply showed the highest tensile strength of 326.40 MPa and carbon-cross 4 ply showed the highest flexural strength of 513.33 MPa. Out of the hybrids, Kevlar-glass cross 4 ply showed the best tensile performance (373.46 MPa). In the prediction, the k-nearest neighbor model was found to be the most robust model (mean squared error = 634.76, mean absolute error = 10.98 and R2 = 0.83), followed by the ensemble random forest model with a balanced performance (R2 = 0.74) and poor performance of the artificial neural network (R2 = 0.41). This work sets up a comprehensive ML system which shows the viability in material selection, which averts the need for extensive experimentation and speeds up composite design.