Abstract: To fully utilize the data in open-source databases and expedite research on inverse spinel oxides, compound data with 1 746 chemical formulas of AB₂O₄ were extracted from the Materials Project (MP) database. From this dataset, 111 inverse spinel oxides were selected, and a feature engineering approach based on cation occupancy was concurrently developed. The resulting data set was used as input variables for machine learning models to predict the formation energies of inverse spinel oxides. Three machine learning models, random forest (RF), gradient boosting regression tree (GBRT), and extreme gradient boosting (XGBoost) were constructed using these algorithms. After performing feature selection, the optimal features influencing the models were identified. Additionally, a comparison was made between the predicted formation energies and the actual values of inverse spinel oxides, the GBRT model that can provide the best predictions was obtained, and which achieves a coefficient of determination (R²) of 0.886 667 and a root-mean-square error (R_MSE) of 0.209 970 on the test set. The trained GBRT model can be utilized to predict the formation energies of inverse spinel oxides, thus expediting the exploration progress of these materials.