An improved container-based deep forest model for predicting groundwater recharge

Abstract This paper proposes ICDF, an improved container-based deep forest model, for effectively modeling and predicting groundwater recharge. The model consists of four points: the construction and expansion of the container module, the assignment of weights to the base model, the growth of the cascade layer, and the decision output. First, container modules are created and a maximization objective function is assigned to each container to control its growth. Next, different weights are assigned to each base model based on its contribution to container prediction. Cascade layers are built using container modules until the model prediction decreases. Finally, the average prediction vectors of the last cascade layer are output. The model’s performance is evaluated and compared with DF and its base models (random forest, adaptive boosting, and extreme gradient boosting) using a case study of 1549 bores in New South Wales, Australia. Remarkably, compared to DF, ICDF has improved prediction accuracy by 6.66%. Moreover, it outperformed the RF, AdaBoost, and XGBoost by 2.94%, 5.85%, and 5.3% in prediction performance, respectively. The ICDF exhibited superior capabilities in predicting groundwater recharge, offering significant improvement over existing models. Practitioners are encouraged to consider adopting ICDF for groundwater management.