Source: https://github.com/aipoch/medical-research-skills

XGBoost Modeling And Feature Importance Ranking

Use this skill to train an XGBoost model from a tabular dataset and export both feature importance ranking tables and feature importance plots.

Use This Skill When

• You need a command-line XGBoost workflow in R for tabular data.
• You need a reproducible train-test split, model training, and evaluation.
• You need feature importance ranking outputs as both a table and a figure.
• You need automatic one-hot encoding for categorical predictors.
• Your data may contain a first unnamed sample ID column such as V1 that should not enter the model.

Do Not Use This Skill When

• Your classification target has more than 2 classes.
• Your input is not tabular CSV, TXT, or TSV data.
• You need causal interpretation, mechanism claims, or policy, business, or clinical conclusions.
• You only need narrative interpretation or triage of an existing result rather than model training.

Primary Command

Rscript scripts/main.R \
  --data_file <input_file> \
  --target_var <target_column> \
  --task_type <auto|classification|regression> \
  --output_dir <output_dir>

Prerequisites

• Rscript is available in the shell.
• Required R packages: optparse, data.table, Matrix, xgboost.
• Install missing packages with Rscript -e 'install.packages(c("optparse", "data.table", "Matrix", "xgboost"), repos="https://cloud.r-project.org")'.

Core Arguments

Argument	Required	Description
--data_file	Yes	Input CSV, TXT, or TSV file
--target_var	Yes	Target column used for modeling
--task_type	No	auto, classification, or regression. Default auto
--output_dir	No	Output directory, default ./XGBoost_Results
--ignore_vars	No	Comma-separated columns to exclude from predictors
--positive_class	No	Positive class label for binary classification
--test_size	No	Test set proportion between 0 and 1, default 0.2
--seed	No	Random seed, default 123
--nrounds	No	Maximum boosting rounds, default 300
--max_depth	No	Tree depth, default 6
--eta	No	Learning rate, default 0.1
--subsample	No	Row sampling ratio, default 0.8
--colsample_bytree	No	Column sampling ratio, default 0.8
--min_child_weight	No	Minimum child weight, default 1
--gamma	No	Minimum split loss reduction, default 0
--lambda	No	L2 regularization, default 1
--alpha	No	L1 regularization, default 0
--early_stopping_rounds	No	Early stopping rounds, default 20
--importance_metric	No	gain, cover, or frequency. Default gain
--top_n	No	Number of features to plot, default 20
--output_format	No	Table format: csv or txt, default csv
--output_prefix	No	Output filename prefix, default xgboost

Input Requirements

• The input file must contain the target column.
• Predictor columns can be numeric, integer, logical, character, or factor-like text.
• Character and factor predictors are one-hot encoded automatically.
• A first unnamed identifier column such as V1 is automatically excluded when it contains unique sample IDs.
• Rows with missing target values are removed before training.
• For classification, exactly 2 classes are required.
• For regression, the target column must be numeric.
• Each class should have at least 2 rows so both training and test sets can be created.

Example input:

,fustat,CAMK2N2,GGT6,GPR161,RAB26,RIBC2
TCGA-C5-A1M5,1,2.248291938,5.274690305,2.825215762,3.121114894,5.35318565
TCGA-EA-A5O9,0,3.346176843,5.404368414,2.604616977,0.629473197,4.429314674
TCGA-C5-A3HL,0,3.363100974,5.363314779,4.124799581,4.127228806,4.916596068

Minimal Workflow

1. Confirm the input file exists and the target column name is correct.
2. Run scripts/main.R with --data_file and --target_var.
3. Check the output directory for table/feature_importance_* and figure/feature_importance_*.

If you omit --data_file or --target_var, the script exits with SKILL_MISSING_INPUT.

Outputs

Expected output structure:

<output_dir>/
├── table/
├── figure/
└── data/

Primary outputs:

• table/<output_prefix>_feature_importance.csv
• table/<output_prefix>_model_performance.csv
• figure/<output_prefix>_feature_importance_<importance_metric>.png

Additional outputs:

• session_info.txt

Feature importance table fields include:

• Rank
• Feature
• Gain
• Cover
• Frequency
• SelectedMetric
• SelectedValue

Feature Importance Metrics

• gain: Average contribution to loss reduction. Recommended for most ranking use cases.
• cover: Relative sample coverage contributed by a feature.
• frequency: How often the feature is used in splits.

Read These Files When Needed

Need	File
XGBoost method details and importance interpretation	references/algorithm.md
More CLI examples	references/cli-guide.md
Error diagnosis	references/troubleshooting.md
Main execution entry point	scripts/main.R
Bundled test data	tests/data/

Quick Examples

Auto-detected binary classification on dt_sample1.csv:

Rscript scripts/main.R \
  --data_file tests/data/dt_sample1.csv \
  --target_var fustat \
  --task_type auto \
  --output_dir tests/output_binary

Binary classification on dt_sample2.csv:

Rscript scripts/main.R \
  --data_file tests/data/dt_sample2.csv \
  --target_var fustat \
  --task_type classification \
  --importance_metric gain \
  --output_dir tests/output_gain

Character-label classification on dt_sample3.txt:

Rscript scripts/main.R \
  --data_file tests/data/dt_sample3.txt \
  --target_var Group \
  --task_type classification \
  --positive_class high \
  --top_n 15 \
  --output_dir tests/output_group

Validation

Rscript scripts/main.R --help

Rscript scripts/main.R \
  --data_file tests/data/dt_sample1.csv \
  --target_var fustat \
  --task_type classification \
  --output_dir tests/validation_output

After running analysis, verify that these files exist:

• tests/validation_output/table/xgboost_feature_importance.csv
• tests/validation_output/table/xgboost_model_performance.csv
• tests/validation_output/figure/xgboost_feature_importance_gain.png

Common Errors

• SKILL_FILE_NOT_FOUND: Input file path is wrong or inaccessible.
• SKILL_MISSING_COLUMNS: The target column is missing.
• SKILL_INVALID_DATA: Data is malformed, the target type is unsuitable, classification has more or fewer than 2 classes, or too few usable rows remain.
• SKILL_INVALID_PARAMETER: An argument value is invalid.
• SKILL_DEPENDENCY_MISSING: A required R package such as xgboost is unavailable.

If the issue is not obvious, read references/troubleshooting.md.