skills.rest
yakeworld
yakeworldCOMMUNITY
Explore all skills by this author →
GitHub

hcs-3wt-breast-cancer-diagnosis

Builds and evaluates a three-stage cascade classifier for breast cancer diagnosis with gray-zone triage.

6|1|Updated May 11, 2026
Search Tags:#cross-validation#scikit-learn#smote#breast cancer diagnosis#cascade classifier#stacking ensemble#three-way decision
One-click install
npx skills add https://github.com/yakeworld/Synthos --skill hcs-3wt-breast-cancer-diagnosis-yakeworld
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: hcs-3wt-breast-cancer-diagnosis
Source: https://github.com/yakeworld/Synthos/tree/main/skills/private/research-tools/research/research-methodology/hcs-3wt-breast-cancer-diagnosis
Command: npx skills add https://github.com/yakeworld/Synthos --skill hcs-3wt-breast-cancer-diagnosis-yakeworld

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires scikit-learn, imbalanced-learn, catboost, numpy, and includes references (resource) components.

What problem does it solve? It provides procedural knowledge for building, evaluating, and optimizing the HCS-3WT (Hybrid Cascade-Stacking Three-Way Triage) system, which automates clear-cut breast cancer diagnoses while routing uncertain cases to a meta-learning arbiter, avoiding the pitfalls of single-model classifiers on clinical data. ## Core Features & Use Cases - Three-Stage Cascade Architecture: Expert B (high-recall SVC) catches clear negatives, Expert A (RF + CatBoost + ExtraTrees voting) confirms clear positives, and Expert C (stacking meta-learner) arbitrates gray-zone cases. - Leakage-Free Evaluation: Enforces cross_val_predict for meta-features, single-class fold skipping, and background execution for 10x5 stratified cross-validation. - Use Case: Run the WDBC experiment (569 samples, 30 features) with fixed thresholds 0.03/0.95 to reproduce ~70.9% automation rate at ~99.3% automation accuracy, or diagnose low-separability datasets like PIMA where fixed thresholds collapse the automation rate. ## Quick Start Run the HCS-3WT pipeline on the WDBC breast cancer dataset with 10x5 cross-validation in the background and report the automation rate, automation accuracy, and gray-zone enrichment.

Frequently Asked Questions about hcs-3wt-breast-cancer-diagnosis

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I build a cascade classifier for breast cancer diagnosis?▼

Use a three-stage design: a high-recall SVC catches clear negatives below a low probability threshold, a voting classifier (RandomForest, CatBoost, ExtraTrees) confirms clear positives above a high threshold, and a stacking meta-learner handles the remaining gray-zone cases.

How to prevent data leakage when training a stacking classifier?▼

Generate the meta-features (P_A and P_B probabilities) fed to the stacking arbiter using cross_val_predict rather than predictions from models trained on the full data. Training meta-features on the same data used for testing creates leakage and inflates results.

Why does Borderline-SMOTE crash during cross-validation?▼

Borderline-SMOTE requires at least two classes in the training data and crashes on single-class folds, as does SVC. Check len(np.unique(y_train)) >= 2 before resampling and skip folds with continue when only one class is present.

Does SVC probability=True still work in sklearn 1.9?▼

No, probability=True is deprecated in sklearn 1.9+. Use CalibratedClassifierCV(SVC(), ensemble=False) instead, with an ImportError fallback to the legacy SVC(probability=True) API for older versions.

Why is my cascade classifier automation rate near zero on new datasets?▼

Fixed thresholds (0.03/0.95) calibrated on breast cancer data fail on low-separability datasets like PIMA, where probabilities rarely reach the extremes. Check probability histograms, then apply dataset-specific threshold optimization or relax thresholds to 0.20/0.80.

Related Skills

Explore More Education & Research Skills→
K-Dense-AIK-Dense-AI
33.0k

pyhealth

Train and evaluate clinical prediction models on EHR datasets.

Official
Advanced
patsnappatsnap
33

precision-oncology-zhcn

Analyze precision oncology questions for specific cancer indications with structured decision support.

Official
Advanced
hung-phanhung-phan
1

evaluation

Evaluate machine learning models on imbalanced datasets with metrics and techniques.

Community
Advanced
awslabsawslabs
13

pa-decision-automation

Parses X12 278 and FHIR PAS transactions into actionable clinical features for adjudication and denial analysis.

Official
Advanced