Changelog

Source: NEWS.md

isoForest 1.1.0

New Features

• Added KDE-Weighted Mean threshold method to set_anomaly_threshold()
  • Density-weighted robust mean using kernel density estimation
  • Automatically weights points by local density for better outlier resistance
  • Particularly effective for heavy-tailed distributions and extreme outliers
  • Added parameter: kde_multiplier
  • Uses KDE to compute density-weighted center of anomaly score distribution
• Added MTT (Modified Thompson Tau Test) threshold method to set_anomaly_threshold()
  • Statistical test based on t-distribution for outlier detection
  • Particularly suitable for small to medium sample sizes (n < 1000)
  • Iterative outlier removal with Bonferroni correction
  • Added parameters: mtt_alpha, mtt_max_iter
  • New internal functions: compute_mtt_threshold(), thompson_tau_critical()
• Enhanced plot_anomaly_boxplot() and plot_anomaly_boxplot_faceted() visualization functions
  • contribution_obj parameter is now optional (defaults to NULL)
  • sample_id can now be a vector to highlight multiple anomalies at once
  • Auto-optimization: point size adjusts to 1 when >5 anomalies are marked
  • Smart subtitle: shows sample IDs for ≤5 anomalies, simplified text for >5
  • Added highlight_size parameter to plot_anomaly_boxplot_faceted()
  • All points (including anomalies) now use jitter effect to avoid overlap
  • Better support for visualizing threshold detection results
• Added plot_anomaly_projection() function for high-dimensional data visualization
  • Supports both PCA and UMAP dimensionality reduction methods
  • Projects high-dimensional data to 2D for intuitive anomaly visualization
  • Anomalies are highlighted in red, normal points in blue
  • Automatic anomaly rate calculation and display
  • PCA: Shows variance explained by principal components
  • UMAP: Configurable parameters (n_neighbors, min_dist)
  • Ideal for exploring anomalies in datasets with >4 features
  • Simple and clean visualization without sample labels
  • Smart sampling based on anomaly rate (default: sample_rate = 0.05)
    • All anomalies are always displayed
    • Normal points are sampled to achieve target anomaly proportion
    • Example: 50 anomalies with 5% rate → ~1000 total points shown
    • Significantly improves speed for large datasets
    • Customizable via sample_rate parameter (set to NULL to disable)
• Added plot_anomaly_projection_all() function for method comparison
  • Side-by-side comparison of PCA and UMAP projections
  • Helps choose the best visualization method for your data
  • Displays both methods with consistent styling
  • Requires gridExtra package for layout

Breaking Changes

• Renamed all plotting functions for better consistency and clarity:
  • anomaly_plot() → plot_anomaly_basic() - Basic 2D anomaly score visualization
  • plot_feature_boxplot() → plot_anomaly_boxplot() - Anomaly boxplot with features
  • plot_feature_boxplot_faceted() → plot_anomaly_boxplot_faceted() - Faceted version
  • plot_anomalies_2d() → plot_anomaly_projection() - Dimensionality reduction projection
  • compare_dim_reduction() → plot_anomaly_projection_all() - Compare PCA vs UMAP
  • All functions now share the unified plot_anomaly_* prefix for easy discovery

Improvements

• Updated mtt_max_iter default value from 10 to 30 for better detection of multiple outliers
• Enhanced documentation with comprehensive method comparison and selection guide
• Added detailed threshold setting section to README.md
• Added “Feature Distribution Visualization” section to README.md with usage examples
• Added “High-Dimensional Data Visualization with Dimensionality Reduction” section to README.md
  • Complete guide for using PCA and UMAP
  • Comparison table and workflow recommendations
  • Advanced usage examples
• Updated vignette with high-dimensional data visualization examples
• Improved plot clarity when displaying many anomaly points simultaneously
• Visualization functions now more flexible for exploring multiple anomalies

isoForest 1.0.0

• Initial CRAN submission.