The production and consumption of non-renewable energy are causes of environmental degradation. To control these effects, a higher percentage of renewable energy should be used in activities instead of non-renewable energy sources. One alternative to solve this problem is to modify the efficiency of generators, converters and storage devices through the use of nanomaterials. Synthesis of these materials in their traditional form requires a large investment of resources and generally a slow yield. However, with methods such as DFT, we can gain access to their physical properties in an accelerated manner. On the other hand, using machine learning (ML), minimal computational resources are needed to make predictions. In this work, ML was used to predict ABO3-type perovskites of the groups Pm-3m, Pnma, P4/mmm, R-3c, I4/mcm, Pbnm, P2-I/c, C2/m, P4mm, Amm2/R3c and Imma, obtaining a total of 1287 compounds. We used as descriptors the formation energy, the energy gap, the polarity (1 if they were not centrosymmetric and 0 if they were), the magnetization and the energy above the contour. Our proposed function for evaluating and prioritizing piezoelectric materials is as follows:

F=P(α·stability +β·norm_band_gap+γ·norm_formation_energy+δ·norm_magnetization

We evaluated four ML models: random forest, XGBoost, gradient boosting and a multilayer perceptron. The best model achieved an R2 value of 0.994 and an RMSE of 0.006; in this case, the best model was XGBoost. According to this model, the three best candidates for piezoelectrics are LaTiO₃, LiNbO₃ and BaTiO₃. With DFT, we calculated the piezoelectric tensor of this material, dij, and the polarization change Pi under deformation, ei.