Random forests are ensemble classifiers and regressors that build hundreds of decision trees in parallel, each on a random bootstrap sample of the training data and with a random subset of features considered at each split, then average their predictions. This combination of bootstrap sampling (bagging) and random feature selection creates diverse, decorrelated trees whose average has much lower variance than any individual tree while preserving the low bias of deep trees.
Random forests are consistently among the top-performing algorithms for tabular data, requiring minimal feature preprocessing, producing reliable feature importances (with known caveats), and being robust to outliers, irrelevant features, and moderate class imbalance. They are often the first complex algorithm to try after logistic regression as a baseline, and their performance frequently sets the benchmark that gradient boosting must beat. Understanding why random forests work — the mechanics of bagging and feature randomisation — explains their robustness and guides effective tuning.