# Algorithm Selection Cheat Sheet

## Start with the problem type

| Problem | Strong first baseline | Alternatives to compare |
|---|---|---|
| Binary classification | Logistic Regression | Decision Tree, Random Forest, SVM |
| Multi-class classification | Multinomial Logistic Regression | Random Forest, Gradient Boosting |
| Numerical prediction | Linear Regression | Tree Regressor, Random Forest, Gradient Boosting |
| Similar groups without labels | k-Means | Hierarchical Clustering, DBSCAN |
| Dimensionality reduction | PCA | UMAP, Autoencoder |
| Anomaly detection | Simple threshold or Isolation Forest | One-Class SVM, Autoencoder |
| Time-series forecasting | Naive Forecast | Exponential Smoothing, ARIMA, boosted trees |
| Recommendation | Popularity or content baseline | Collaborative Filtering, Matrix Factorization |
| Text classification | Logistic Regression with TF-IDF | Naive Bayes, Transformer Classifier |
| Image classification | Transfer-learned CNN | Vision Transformer |

## Selection questions

1. Is the target a category, number, group, order, sequence or generated output?
2. Are labels available and trustworthy?
3. How large and high-dimensional is the dataset?
4. Is interpretation required?
5. What latency, memory and cost limits exist?
6. What errors create the greatest harm?

Never select an algorithm by popularity alone. Compare a baseline using relevant evaluation evidence.