RFOD: Random Forest-Based Outlier Detection for Mixed-Type Tabular Data

Outlier detection in tabular data is crucial for safeguarding data integrity in high-stakes domains such as cybersecurity, financial fraud detection, and healthcare, where anomalies can lead to severe operational and economic risks. Despite advances in both data mining and deep learning, existing methods often struggle with mixed-type tabular data, where numerical and categorical features coexist with complex dependencies and skewed distributions. They typically rely on global heuristics that overlook context-dependent irregularities, while deep neural models require heavy preprocessing and sacrifice interpretability. To overcome these challenges, we introduce RFOD, a Random Forest-based Outlier Detection framework that reformulates anomaly detection as a series of feature-wise conditional reconstruction tasks. Each random forest predicts a feature conditioned on the others, capturing localized dependencies and uncovering both global and subsetspecific anomalies without explicit context definition. For precise and interpretable detection, RFOD integrates adjusted Gower's distance (AGD), which adapts to skewed numerical data and accounts for categorical confidence, with uncertainty-weighted averaging (UWA) to aggregate cell-level scores into robust rowlevel anomaly scores. Extensive experiments on 15 real-world datasets show that RFOD consistently outperforms ten stateof-the-art baselines, achieving an average AUC-ROC gain of 16.9% over data-mining methods and 25.7% over deep-learning approaches, while maintaining superior robustness, scalability, and interpretability on mixed-type tabular data. Code is available at https://github.com/YihaoAng/RFOD/.