What problem does it solve?
Choosing the right predictors for a classification or regression model is often ad-hoc and error-prone. This Skill replaces hand-picked feature sets with a structured, data-driven MATLAB pipeline that generates a candidate feature pool, ranks it with a consensus of multiple rankers, assesses the result, and delivers a reproducible inference script.
Core Features & Use Cases
- Automated feature generation: Builds a candidate pool with gencfeatures/genrfeatures from the Statistics and Machine Learning Toolbox, or routes non-tabular data (signals, images, telemetry) to a matching domain extraction skill.
- Consensus feature selection: Runs a ranker panel, a consensus vote, and an elbow cut on training rows only, with strict train/test discipline to prevent leakage.
- Assessment and reporting: Scores the selected set against raw and naive baselines (holdout or cross-validated), offers optional stability gates, and produces a markdown/PDF report plus a self-contained fe_transform inference function.
- Use Case: Given a patient records table, ask the agent to engineer and select the best features for predicting readmission; it screens degenerate columns, builds and ranks the pool, validates the selection, and delivers a reusable transform script.
Quick Start
Ask your AI agent to engineer and select the best predictive features from your MATLAB table for a classification or regression target, and confirm the output directory when prompted.
