CloudForest

Achieves faster training times than many implementations due to CPU cache-friendly memory utilization and separate, optimized paths for binary, numerical, and categorical data, as shown in benchmarks on heterogeneous clinical data.

Includes bias correction and three-way splitting by default to handle missing data without imputation, reducing bias towards features with lots of missing values as described in the Missing Values section.

Implements roughly balanced bagging, cost-weighted classification, and weighted Gini impurity for predicting rare events in unbalanced classes, which is critical for genetic and clinical studies.

Offers artificial contrasts with ensembles (ACE) for improved feature selection with p-values, helping combat overfitting in noisy, high-dimensional data as detailed in the Importance section.