About this skill

Problem

Structured-data analysis tasks often break apart after the model is trained: EDA, evaluation, explanation, and paper writing are handled separately. Metrics may be produced, but figures lack business interpretation, formulas remain plain text, leakage is not audited, and the output is not ready for academic or competition submission.

Workflow

The skill links mathematical modeling, machine learning, and paper writing into a fixed pipeline:
- EDA: inspect distributions, skew, and correlations, then produce feature distributions, segment shares, segment comparisons, and correlation heatmaps.
- Modeling: support XGBoost, regression, classification, and clustering, with baselines such as RandomForest, and task-specific metrics like AUC, F1, RMSE, or silhouette score.
- Explainability: emit feature importance, optional SHAP summaries, or model comparisons instead of reporting one model in isolation.
- Paper standards: use academic or CUMCM competition mode, with LaTeX formulas, symbol definitions, GB/T 7714 references, and figure interpretations.

Boundaries

It is intended for structured data such as user behavior, transactions, sensors, or experiments. If a test set reaches AUC=1 or R^2=1, treat that as a leakage signal first. Non-OpenClaw environments can output local Markdown and figures; real projects should replace sample data with business data and add confidence intervals or significance tests.

Screenshots

Use Cases

Build RFM segments and an XGBoost churn model from e-commerce behavior tables, then produce a paper with charts and formulas.
Prepare a CUMCM competition paper with abstract, assumptions, model derivation, solution, and sensitivity checks.
Turn sensor experiment data into EDA, regression modeling, and feature-importance analysis for a methods section.
Train a churn classifier on transaction or behavior data and generate AUC, F1, SHAP, and result discussion.

Best For

Data analysts who need to turn structured business data into EDA, modeling, and reportable conclusions.
Mathematical modeling contestants who need a CUMCM-style paper with required sections and formatting.
Graduate students or researchers who need to write machine learning experiments, formulas, and references into an academic draft.
Algorithm engineers who need baseline comparisons, leakage checks, and interpretability charts to support model explanations.