Explainable AI Analysis

Comprehensive explainable AI toolkit for interpreting machine learning models in medical and clinical applications. Provides attention maps, feature importance, SHAP values, and other interpretability methods.

Usage

explainableai(
  data,
  analysis_type = "feature_importance",
  features,
  target_var,
  model_predictions,
  image_paths,
  attention_maps,
  shap_method = "kernel_explainer",
  lime_method = "tabular",
  n_samples = 100,
  n_features = 20,
  plot_type = "summary",
  overlay_original = TRUE,
  confidence_level = 0.95,
  background_samples = 100,
  perturbation_method = "random",
  clustering_method = "none",
  interaction_analysis = FALSE,
  local_explanations = TRUE,
  global_explanations = TRUE,
  save_explanations = FALSE,
  explanation_path = "",
  attention_threshold = 0.1,
  colormap = "viridis"
)

Arguments

data

the data as a data frame

analysis_type

type of explainability analysis to perform

features

features/variables for importance analysis

target_var

target variable or model predictions

model_predictions

variable containing model predictions or probabilities

image_paths

variable containing paths to image files

attention_maps

variable containing attention map data or file paths

shap_method

SHAP explainer method based on model type

lime_method

LIME explanation method for different data types

n_samples

number of samples to use for explanation analysis

n_features

number of top important features to display

plot_type

type of visualization for explanations

overlay_original

overlay attention maps on original images

confidence_level

confidence level for statistical intervals

background_samples

number of background samples for SHAP baseline

perturbation_method

method for perturbing features in permutation importance

clustering_method

method for grouping similar features

interaction_analysis

analyze pairwise feature interactions

local_explanations

create individual sample explanations

global_explanations

create overall model explanations

save_explanations

save explanation data for external use

explanation_path

file path to save explanation results

attention_threshold

minimum attention value to display in visualizations

colormap

color palette for explanation visualizations

Value

A results object containing:

results$overview					a html
results$featureimportance$importancetable					Ranked list of feature importance scores
results$featureimportance$importanceplot					Bar plot of feature importance scores
results$shapanalysis$shapvaluestable					Mean absolute SHAP values per feature
results$shapanalysis$shapwaterfalltable					Individual prediction explanations
results$shapanalysis$shapinteractiontable					Feature interaction effects
results$shapanalysis$shapsummaryplot					Overview of SHAP values for all features
results$shapanalysis$shapwaterfallplot					Individual prediction explanation
results$shapanalysis$shapinteractionplot					Feature interaction heatmap
results$limeanalysis$limeexplanationtable					Local explanations for individual predictions
results$limeanalysis$limeplot					Local explanation visualization
results$attentionanalysis$attentionstatstable					Summary statistics of attention patterns
results$attentionanalysis$attentionpeakstable					Highest attention regions across samples
results$attentionanalysis$attentionheatmapplot					Attention map overlays on original images
results$attentionanalysis$attentiondistributionplot					Distribution of attention values
results$partialdependence$pdptable					Partial dependence effect sizes
results$partialdependence$pdpplot					Effect of individual features on predictions
results$partialdependence$iceplot					Individual conditional expectation curves
results$globalexplanations$modelinsightstable					Overall model behavior insights
results$globalexplanations$featureclusteringtable					Groups of similar features
results$globalexplanations$globalinsightplot					Overall model behavior visualization
results$globalexplanations$featureclusterplot					Dendrogram or cluster plot of features
results$localexplanations$samplewiseexplanationtable					Detailed explanations for individual samples
results$localexplanations$localexplanationplot					Individual sample explanation examples
results$validationmetrics$validationtable					Quality metrics for explanations
results$validationmetrics$stabilityanalysistable					Consistency of explanations across perturbations
results$validationmetrics$validationplot					Validation metrics visualization

Examples

# SHAP analysis for feature importance
explainableai(
    data = model_data,
    model_predictions = "predictions_var",
    features = c("feature1", "feature2", "feature3"),
    method = "shap",
    plot_type = "summary"
)

# Attention map analysis for image models
explainableai(
    data = image_data,
    image_paths = "image_path_var",
    attention_maps = "attention_var",
    method = "attention_analysis",
    overlay_original = TRUE
)