What problem does it solve?
Choosing the right classifier for a tabular dataset usually means guessing among dozens of MATLAB functions and then wondering whether accuracy differences are real or noise. This Skill runs an end-to-end MATLAB workflow that cleans your data, selects suitable classifiers based on dataset characteristics, trains them, and uses statistical tests (McNemar, 5x2 cv, Friedman) to identify which models are genuinely equivalent in accuracy.
Core Features & Use Cases
- Data-driven model selection: Analyzes your dataset (size, class balance, missing values, categorical features, sparsity) and picks appropriate classifiers such as trees, SVMs, ensembles, naive Bayes, and neural networks.
- Rigorous evaluation: Supports 5-fold cross-validation or 70/30 holdout splits, with confidence intervals, pairwise p-value matrices, Bonferroni correction, and heatmap visualizations.
- Imbalanced data handling: Offers RUSBoost and other imbalance-aware boosting models plus an optional uniform class prior.
- Optional extras: Hyperparameter optimization via Bayesian optimization, boosting learning curves with resume training, saving trained models to .mat files, and exporting a self-contained retraining script.
- Use Case: You have a CSV of patient measurements and a diagnosis column. Ask your agent to classify the data, and it will clean the table, train a dozen candidate models, and tell you which top-tier models are statistically indistinguishable so you can pick the simplest one.
Quick Start
Ask your AI agent to classify the tabular dataset in my CSV file and compare the accuracy of different MATLAB classifiers with cross-validation.
