Combined Pruning for Nested Cross-Validation to Accelerate Automated Hyperparameter Optimization for Embedded Feature Selection in High-Dimensional Data with Very Small Sample Sizes

Applying tree-based embedded feature selection to exclude irrelevant features in high-dimensional data with very small sample sizes requires optimized hyperparameters for the model building process. In addition, nested cross-validation must be applied for this type of data to avoid biased model performance. The resulting long computation time can be accelerated with pruning. However, standard pruning algorithms must prune late or risk aborting calculations of promising hyperparameter sets due to high variance in the performance evaluation metric. To address this, we adapt the usage of a state-of-the-art successive halving pruner and combine it with two new pruning strategies based on domain or prior knowledge. One additional pruning strategy immediately stops the computation of trials with semantically meaningless results for the selected hyperparameter combinations. The other is an extrapolating threshold pruning strategy suitable for nested-cross-validation with high variance. Our proposed combined three-layer pruner keeps promising trials while reducing the number of models to be built by up to 81,3% compared to using a state-of-the-art asynchronous successive halving pruner alone. Our three-layer pruner implementation(available at https://github.com/sigrun-may/cv-pruner) speeds up data analysis or enables deeper hyperparameter search within the same computation time. It consequently saves time, money and energy, reducing the CO2 footprint.