Works with the AI Agents you already use
classification model development
Builds a binary classifier from an approved labelled dataset: leakage and consent screening, holdout precision, recall and calibration, cost-based threshold, group evaluation, model card. Decision support only.
$2.99
classification model development
Example session with this skill installed
Develop a binary classifier for 'loan_default' using E-122 data. Target approved by Sarah Miller. Costs: FP 500, FN 2000. Features in loan_data.csv. Ensure calibration is reported.
- Read your context and instructions
- Compiled the classification model development
The classification model for `loan_default` has been developed using the deterministic pipeline. All records without explicit consent were excluded prior to training. # Classification Model Development — loan_default, 2023-10-24 ## 1. Scope, sources and assumptions - **Sources**: E-122 (Loan application records) - **Approvals**: Target and dataset approved by Sarah Miller. - **Method**: Stratified split, seed 42. ## 2. Data, split and class balance | Metric | Value | | :--- | :--- | | Records Supplied | 10,000 | | Records Excluded (No Consent) | 150 | | Training Count | 6,895 | | Holdout Count | 2,955 | | Imbalance Ratio | 1:8.4 | ## 3. Features and leakage screening - `income`: Numeric, included. - `credit_score`: Numeric, included. - `current_loan_status`: Excluded (Available only after outcome). ## 4. Model selection and artifact - **Selected Model**: Gradient Boosted Tree - **OOF Log Loss**: 0.241 - **Training Hash**: `sha256:7b9a...` ## 5. Threshold analysis - **Selected Threshold**: 0.32 - **Method**: Cost-minimization (FP: 500, FN: 2000) - **Constraints**: Met (Min Recall > 0.70) ## 6. Holdout evaluation and calibration | Metric | Estimate | 95% CI | | :--- | :--- | :--- | | Precision | 0.682 | [0.651, 0.713] | | Recall | 0.741 | [0.710, 0.772] | | ROC AUC | 0.884 | [0.865, 0.903] | | ECE | 0.012 | N/A | ## 7. Group evaluation Recall gap between `region_A` and `region_B` is 0.04. Selection rate ratio is 0.92. ## 8. Warnings and items for human review - **Information gap — organizational validation required**: Confirm cost weights represent current fiscal year risk. - **Analytical estimate — validation required**: Threshold selection assumes stable default rates. ## 9. Model card summary and approval boundary Model exhibits strong calibration (ECE < 0.02). Adopting the 0.32 threshold is a **Management decision required**. This analysis supports, and does not replace, professional judgment and management's decision. **Next steps** 1. Review the `model_artifact.bin` for integration testing. 2. Validate the cost-ratio assumptions with the finance team. 3. Approve the model card for the internal compliance registry.
Connects securely to your tools. The creator never sees your data.
About this skill
Manual classification modeling is prone to data leakage, arbitrary threshold selection, and poor calibration. This skill automates the development of a binary classifier through a deterministic, reproducible pipeline that enforces rigorous holdout evaluation. ## What it does - **Leakage screening** identifies and excludes features that would not be available at the time of prediction. - **Automated model selection** evaluates logistic or tree-based models using out-of-fold log-loss and Brier scores. - **Cost-based thresholding** selects decision boundaries based on specific costs of false positives versus false negatives. - **Fairness reporting** calculates recall gaps and selection-rate ratios across defined groups to highlight disparities. - **Model card generation** produces a technical summary including training hashes, coefficients, and limitations for audit. ## How it works 1. The developer provides a labeled dataset, target definition, and explicit approval/consent markers. 2. The agent builds a `request.json` configuration, mapping features to their types and availability. 3. A deterministic execution script (`run.py`) performs the splitting, training, and evaluation. 4. The agent interprets the resulting envelope, reporting metrics, calibration curves, and any triggered warnings. ## Frameworks & tools The skill utilizes a Python-based execution environment (`python3`) to run a specialized `scripts/run.py` utility. It outputs structured Markdown reports based on `input.schema.json` and `output.schema.json` definitions. ## Why this beats prompting it yourself Generic prompts often lead to "lazy" evaluation where models overfit on the entire dataset or ignore class imbalance. This skill enforces a strict temporal or stratified split and requires explicit cost-benefit inputs for thresholding, ensuring the model solves the actual business problem rather than just maximizing accuracy. ## Use cases - Build a churn prediction model with a threshold optimized for the cost of retention offers. - Evaluate a fraud detection flag while screening for data leakage and group bias. - Generate a reproducible model card and artifact for a risk assessment classifier. - Perform a calibration check on an existing classification target to ensure predicted probabilities match real-world outcomes. ## Known limitations This skill is for binary classification only and does not support multi-class or regression tasks. It requires pre-approved datasets with explicit consent markers and cannot be used for causal inference or time-series forecasting.
Binary classification only. No multi-class or regression. Requires pre-approved data and explicit consent markers. No causal inference or live deployment.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 3 days ago