An Empirical Comparison of Voting Classification Algorithms: Bagging, Boosting, and Variants
E. BauerRon Kohavi
Demonstrates through empirical bias-variance decomposition how bagging and boosting alter classification error, showing that bagging primarily reduces variance in unstable models while boosting reduces both bias and variance but struggles with noisy data and stable learners like Naive-Bayes.
The article addresses the challenge of improving classification accuracy using voting methods such as Bagging and boosting algorithms like AdaBoost. These techniques have demonstrated success on both artificial and real-world datasets, yet the reasons they reduce error for certain inducers remain unclear, particularly regarding their impact on bias and variance. Understanding these mechanisms matters because accurate classifiers affect practical decisions in many domains.
The article set out to evaluate Bagging, AdaBoost, Arc-x4, and several variants in conjunction with decision tree inducers and a Naive-Bayes inducer. It aimed to determine why and when these perturbation, reweighting, and combination methods lower classification error.
The authors conducted a large empirical study across 14 datasets, each with at least 1,000 instances. They applied bias-variance decomposition to isolate effects, tested variants including pruning choices, probabilistic estimates, and backfitting, and performed sanity checks against prior experiments. Training and test splits followed learning-curve guidance to ensure room for improvement.
Bagging reduced variance for unstable inducers like decision trees, cutting average error from 12.6 percent to 10.4 percent, while boosting methods lowered both bias and variance for the same inducers, achieving a 27 percent relative error reduction. Boosting increased variance for the stable Naive-Bayes inducer yet still improved overall accuracy. Arc-x4 performed comparably to AdaBoost when resampling was used but worse with reweighting alone. No-pruning combined with probabilistic estimates further improved Bagging, and mean-squared error dropped substantially across voting methods. Boosting sometimes raised error on noisy datasets.
These findings indicate that voting methods deliver meaningful gains in accuracy and calibration where comprehensibility is secondary. They suggest practitioners should prefer boosting or refined Bagging variants over single trees or Naive-Bayes for many tasks, though gains vary by data noise and inducer stability.
The article recommends exploring noise-robust boosting variants, parallel implementations, and methods to preserve interpretability. Additional work is needed on tree-size dynamics and handling zero-error trials during boosting.
The study is limited to two inducer families and 14 datasets; results may not generalize to other algorithms or noisier domains. Confidence is moderate given the controlled experimental design, yet caution is warranted for noisy data or when many trials are required.
- Paper: Bagging Predictors, L. Breiman (1996). This seminal paper introduces the bagging predictor algorithm and its variance-reduction mechanism, which serves as a core ensemble method evaluated in the benchmark.
- Paper: A Decision-Theoretic Generalization of On-Line Learning and an Application to Boosting, Yoav Freund et al. (1997). This foundational work establishes the theoretical formulation of AdaBoost, the central boosting algorithm evaluated throughout the study.
- Paper: Experiments with a New Boosting Algorithm, Yoav Freund et al. (1996). This paper presents the early empirical comparisons of AdaBoost and bagging across benchmark datasets, forming the experimental precedent directly built upon by the source.
- Paper: Boosting the margin: A new explanation for the effectiveness of voting methods, R. Schapire et al. (1997). This work introduces the margin-based theoretical explanation for why boosting and voting ensembles reduce generalization error.
- Paper: Induction of Decision Trees, J. R. Quinlan (1986). This paper establishes the foundational decision-tree induction methodology that forms the primary base learner in the source's empirical study.
- Paper: The Random Subspace Method for Constructing Decision Forests, Tin Kam Ho (1998). This work introduces the random subspace method for building tree ensembles, providing essential context on feature perturbation alongside bootstrap aggregating.
- Paper: A Study of Cross-Validation and Bootstrap for Accuracy Estimation and Model Selection, Ron Kohavi (1995). This study provides foundational methodology for cross-validation and bootstrap accuracy estimation applied in large-scale empirical machine learning benchmarks.
- Paper: Approximate Statistical Tests for Comparing Supervised Classification Learning Algorithms, Thomas G. Dietterich (1998). This paper details statistical testing protocols for comparing supervised classification algorithms, framing the evaluation design used in the study.
- Paper: An Experimental Comparison of Three Methods for Constructing Ensembles of Decision Trees: Bagging, Boosting, and Randomization, Thomas G. Dietterich (2000). Dietterich builds upon empirical comparisons of bagging and boosting by systematically investigating randomized tree construction and robustness under synthetic label noise.
- Paper: Greedy function approximation: A gradient boosting machine, Jerome H. Friedman (2001). Friedman generalizes boosting beyond classification into a functional gradient descent framework for arbitrary differentiable loss functions.
- Paper: Random Forests, Leo Breiman (2001). Breiman extends bagging by combining bootstrap sampling with random feature selection to introduce the Random Forest ensemble algorithm.
- Paper: An empirical comparison of supervised learning algorithms, R. Caruana et al. (2006). This large-scale benchmark expands comparative evaluation beyond tree ensembles to modern classifiers and probability calibration metrics.
- Paper: Extremely randomized trees, Pierre Geurts et al. (2006). This work advances randomized ensemble methods by introducing extremely randomized trees to further optimize the bias-variance trade-off and computational efficiency.
- Paper: Predicting good probabilities with supervised learning, Alexandru Niculescu-Mizil et al. (2005). This study examines the probability calibration distortions identified in boosted and bagged models and proposes post-processing correction methods.
- Paper: XGBoost: A Scalable Tree Boosting System, Tianqi Chen et al. (2016). Chen and Guestrin build on gradient boosting principles to develop XGBoost, an ultra-scalable and regularized tree boosting system.
- Paper: CatBoost: unbiased boosting with categorical features, Liudmila Prokhorenkova et al. (2018). CatBoost extends modern gradient boosting to address target leakage and improve handling of categorical features.
- Paper: Statistical Comparisons of Classifiers over Multiple Data Sets, Janez Demšar (2006). Demšar provides formal non-parametric statistical testing frameworks specifically for comparing classifier performance across multiple benchmark datasets.
- Paper: A comparative analysis of gradient boosting algorithms, Candice Bentéjac et al. (2019). This paper presents a modern comparative evaluation of second-generation gradient tree boosting implementations including XGBoost, LightGBM, and CatBoost.
