GLIME: A new graphical methodology for interpretable model-agnostic explanations

Beyond single features: gLIME reveals how features work together — While traditional explainability tools like LIME only show which individual features matter most, gLIME goes further by displaying the relationships and dependencies between features, helping users understand how multiple factors combine to influence a model's decision.

Two levels of explanation for different needs — The tool can provide explanations at both global scale (showing patterns across an entire dataset) and local scale (explaining specific individual predictions), making it useful for different types of questions about model behavior.

Captures both direct and indirect impacts — gLIME distinguishes between features that directly influence a model's output and those that have indirect effects through their relationships with other features, providing a more complete picture of how decisions are made.

Works with any type of model — As a "model-agnostic" tool, gLIME can be applied to any black box AI system regardless of its underlying architecture, making it broadly applicable across different industries and use cases.

Tested on real medical datasets — The methodology was validated using established classification datasets (BIOPSY and OAI), demonstrating its practical applicability to real-world problems where understanding AI decisions is critical, such as healthcare.