Mechanical property predictions in multi-material lattice structures are very important in order to optimize lightweight designs in the field of engineering and biomedical applications. In this work, the behaviour of gyroid structures made of thermoplastic polyurethane (TPU) and polylactic acid (PLA) was predicted using machine learning (ML) algorithms. This study incorporates two key design parameters (PLA percentage and relative density) of gyroid structures. Since TPU content is complementary to PLA, it was not used as an independent variable. Three mechanical properties were assessed as target outputs: elastic modulus (EM), energy absorption (EA), and peak stress (PS). A systematic comparison of five ML algorithms (artificial neural networks, support vector regression, random forest, decision tree, and CatBoost) was carried out using the mean absolute error (MAE), root mean squared error (RMSE), and mean arctangent absolute percentage error (MAAPE) for each property. The results indicated that CatBoost maintained consistently well-balanced predictive performance. Meanwhile, random forest exhibited the highest robustness and scale-independent performance across all mechanical properties.