Overview of core ML models used across projects, why each is chosen, and the types of insights you can extract from them.
Why: Fast, regularized gradient boosting with strong performance on tabular data and good handling of missing values.
Insights: Feature importance, SHAP values for per-prediction explanations, partial dependence to show marginal feature effects.
Why: Highly optimized gradient-boosting implementation that trains quickly on large datasets and supports categorical features efficiently.
Insights: Fast experiments for hyperparameter tuning, feature importance, and sub-sample analyses for large-scale pipelines.
Why: Robust ensemble baseline that reduces variance; useful where interpretability and stability matter.
Insights: Out-of-bag error estimates, permutation feature importance, and per-tree diagnostics to understand variance.
Why: Flexible models for learning complex, non-linear relationships; useful for time-series, embeddings, or when combining many inputs.
Insights: Layer activations and embeddings, saliency maps, and sequence-level attention weights for interpretability.
Why: Combine strengths of multiple algorithms to improve robustness and accuracy, reduce model-specific biases.
Insights: Consensus/confidence across models, model-level contribution analysis, and ensemble calibration diagnostics.
If you want, I can add links to example notebooks, SHAP plots, or a `requirements.txt` for each model.