Build ML Models with Scikit-learn

Scikit-learn is a skill providing comprehensive guidance for classical machine learning with the industry-standard Python library - classification, regression, clustering, dimensionality reduction, preprocessing, model evaluation, and production-ready ML pipelines. Installation is via uv pip install scikit-learn, with matplotlib/seaborn for visualization and pandas/numpy commonly paired alongside it.

Supervised learning covers linear models (Logistic/Linear Regression, Ridge, Lasso, ElasticNet), tree-based methods (Decision Trees, Random Forest, Gradient Boosting), Support Vector Machines with various kernels, ensemble methods (AdaBoost, Voting, Stacking), neural networks (MLPClassifier/MLPRegressor), and Naive Bayes/KNN - for classification (discrete categories like spam or fraud detection) and regression (continuous values like price or demand). Unsupervised learning covers clustering (K-Means, MiniBatchKMeans, DBSCAN, HDBSCAN, OPTICS, AgglomerativeClustering, Gaussian Mixture, MeanShift, SpectralClustering, BIRCH) and dimensionality reduction (PCA, TruncatedSVD, NMF for linear methods; t-SNE, UMAP, Isomap, LLE for manifold learning; FastICA and LDA for feature extraction), used for customer segmentation, anomaly detection, visualization, and topic modeling.

Model evaluation and selection covers cross-validation strategies (KFold, StratifiedKFold, TimeSeriesSplit for temporal data, GroupKFold for grouped samples), hyperparameter tuning (GridSearchCV, RandomizedSearchCV, HalvingGridSearchCV), and metrics by task - classification (accuracy, precision, recall, F1, ROC AUC, confusion matrix), regression (MSE, RMSE, MAE, R², MAPE), and clustering (silhouette score, Calinski-Harabasz, Davies-Bouldin). Data preprocessing covers scaling (StandardScaler, MinMaxScaler, RobustScaler, Normalizer), categorical encoding (OneHotEncoder, OrdinalEncoder, LabelEncoder), missing value handling (SimpleImputer, KNNImputer, IterativeImputer), and feature engineering (PolynomialFeatures, KBinsDiscretizer, RFE/SelectKBest/SelectFromModel for feature selection).

Production workflow guidance centers on Pipeline/ColumnTransformer/FeatureUnion/TransformedTargetRegressor composition - chaining transformers and estimators, applying different preprocessing per column type, and combining transformers in parallel - which prevents data leakage during cross-validation, simplifies maintenance, enables joint hyperparameter tuning across preprocessing and model steps, and ensures training/prediction consistency. Two bundled example scripts run complete workflows end to end: classification_pipeline.py (mixed data type handling, cross-validated model comparison, GridSearchCV tuning, multi-metric evaluation, feature importance) and clustering_analysis.py (optimal cluster count via elbow method and silhouette analysis, comparing K-Means/DBSCAN/Agglomerative/Gaussian Mixture, PCA-projected visualization). Reference documentation is organized into quick_reference, supervised_learning, unsupervised_learning, model_evaluation, preprocessing, and pipelines_and_composition files for deep dives.

Use this skill when building classification or regression models, performing clustering or dimensionality reduction, preprocessing data for ML, evaluating model performance with cross-validation, tuning hyperparameters, building production ML pipelines, comparing algorithms, or needing interpretable classical ML on structured/tabular data.

Do not use it for deep learning tasks requiring GPU-accelerated neural network training at scale - scikit-learn's MLP implementations are CPU-based and suited to smaller networks; for large-scale deep learning, a framework like PyTorch or TensorFlow is the right tool instead.

Input: labeled or unlabeled tabular (or text) data for a classification, regression, clustering, or dimensionality-reduction task.

Output: a trained model or pipeline, for example a complete mixed-data classification pipeline:

numeric_transformer = Pipeline([
    ('imputer', SimpleImputer(strategy='median')),
    ('scaler', StandardScaler())
])

categorical_transformer = Pipeline([
    ('imputer', SimpleImputer(strategy='most_frequent')),
    ('onehot', OneHotEncoder(handle_unknown='ignore'))
])

preprocessor = ColumnTransformer([
    ('num', numeric_transformer, numeric_features),
    ('cat', categorical_transformer, categorical_features)
])

model = Pipeline([
    ('preprocessor', preprocessor),
    ('classifier', GradientBoostingClassifier(random_state=42))
])

model.fit(X_train, y_train)
y_pred = model.predict(X_test)

Along with this it produces cross-validation results, hyperparameter search results, and clustering/dimensionality-reduction analysis with visualizations.

Data scientists and ML engineers working with structured/tabular data who need comprehensive, production-oriented scikit-learn guidance rather than piecing together preprocessing, modeling, and evaluation code from scattered documentation.