Explainable AI (XAI): Making Artificial Intelligence Transparent

Abstract

Deep learning and ensemble models now drive decisions in healthcare, finance, and judicial systems, yet their internal reasoning remains largely opaque to the humans who rely on them. This opacity — the \"black-box problem\" — limits trust, complicates regulatory compliance, and makes it difficult to detect bias or error before harm occurs. Explainable Artificial Intelligence (XAI) addresses this gap by producing human-understandable justifications for model predictions. This paper reviews the major families of XAI techniques — intrinsically interpretable models, and post-hoc methods such as LIME, SHAP, Grad-CAM, counterfactual explanations, and attention-based interpretability — and proposes a hybrid explainability framework that combines global feature attribution (SHAP), local instance-level explanation (LIME), and rule-based surrogate extraction, evaluated through a novel Explanation Consistency Index (ECI). The framework is demonstrated on the Wisconsin Diagnostic Breast Cancer (WDBC) dataset. The goal is to show that transparency and predictive performance need not be traded off against one another, and that hybrid explanation strategies can improve both trust and auditability of machine learning systems.

Introduction

The paper proposes a Hybrid Explainable Artificial Intelligence (XAI) Framework that aims to make machine learning models more transparent, understandable and trustworthy.

Artificial intelligence models, especially deep learning models, can achieve high prediction accuracy, but their decision-making processes are often difficult to understand. This creates problems in sensitive areas such as healthcare, banking, credit scoring and criminal risk assessment, where users need to know why a particular decision was made.

The proposed framework combines multiple explainability techniques to explain machine learning predictions and measure the consistency of their explanations.

1. Introduction and Need for Explainability

Traditional machine learning models often work like black boxes because they produce predictions without clearly explaining how they reached those decisions.

Explainable AI (XAI) helps solve this problem by:

• Explaining why a model made a particular prediction.

• Helping developers identify model errors.

• Detecting possible bias and unfair decisions.

• Improving transparency and user trust.

• Supporting auditing and regulatory requirements.

• Helping users understand AI-based decisions.

2. Types of Explainable AI Techniques

XAI techniques can be classified according to three categories:

• Intrinsic vs. Post-hoc: Whether interpretability is built into the model or added after training.

• Model-specific vs. Model-agnostic: Whether the explanation method works with a particular model or different model types.

• Local vs. Global: Whether it explains an individual prediction or the overall model behaviour.

The paper discusses five major explainability techniques:

A. LIME (Local Interpretable Model-Agnostic Explanations)

• Explains individual predictions.

• Creates a simple interpretable model around a particular prediction.

• Identifies the features that influence the decision.

• Works with different machine learning models.

B. SHAP (SHapley Additive exPlanations)

• Calculates the contribution of individual features to a prediction.

• Uses cooperative game theory.

• Provides both local and global explanations.

• Helps rank the most influential features.

C. Grad-CAM and Saliency Maps

• Mainly used for image-based deep learning models.

• Highlights image regions that influence a prediction.

• Useful in medical imaging and object recognition.

D. Counterfactual Explanations

• Explain what changes in input data could change a model's prediction.

• For example, explain what changes in a loan applicant's information might lead to loan approval.

E. Attention-Based Interpretability

• Uses attention weights from transformer models.

• Highlights important words or parts of input text.

• Helps understand how transformer models process information.

3. Related Work

Previous researchers have developed different XAI techniques:

• Ribeiro et al. introduced LIME for explaining individual predictions.

• Lundberg and Lee introduced SHAP for feature contribution analysis.

• Selvaraju et al. proposed Grad-CAM for visual explanations.

• Tjoa and Guan studied explainability in medical applications.

• Bussmann et al. applied SHAP to financial risk models.

• Rudin highlighted the importance of inherently interpretable models in high-risk applications.

• Doshi-Velez and Kim proposed frameworks for evaluating explainability.

These studies establish the importance of explainability but also highlight limitations such as unstable explanations, computational costs and difficulty in verifying whether explanations truly represent model behaviour.

4. Problem Statement and Research Gap

High-performing machine learning models often provide accurate predictions without explaining the reasons behind them.

Existing explanation techniques have individual limitations:

• LIME may produce unstable explanations.

• SHAP can be computationally expensive.

• Grad-CAM is mainly designed for image-based models.

• Counterfactual explanations may suggest unrealistic changes.

• Attention weights do not always represent the actual reasoning process.

The proposed research addresses this problem by combining multiple explanation methods and introducing a measurement to evaluate how consistently they explain the same model.

5. Proposed System and Methodology

The framework uses the Wisconsin Diagnostic Breast Cancer (WDBC) dataset to demonstrate the proposed methodology.

The dataset contains:

• 569 total samples.

• 30 numerical features.

• Two classification categories: Benign and Malignant.

The methodology consists of the following steps:

1. Data Collection: Load the WDBC breast cancer dataset.

2. Data Preprocessing: Standardize numerical features and handle missing values if present.

3. Dataset Splitting: Divide the dataset into 80% training data and 20% testing data while maintaining class balance.

4. Model Training: Train a Random Forest classifier as the black-box predictive model.

5. Global Explanation: Apply SHAP to identify the most influential features across the dataset.

6. Local Explanation: Apply LIME to explain individual predictions.

7. Rule Extraction: Train a shallow decision tree to generate understandable if-then rules that approximate the black-box model.

8. Consistency Evaluation: Calculate the Explanation Consistency Index (ECI).

9. Performance Evaluation: Evaluate prediction accuracy and explanation quality.

6. Proposed Hybrid Explainability Framework

The main contribution is combining three explanation techniques:

These methods provide different perspectives on the same machine learning model.

7. Explanation Consistency Index (ECI)

The paper introduces a proposed metric called the Explanation Consistency Index (ECI).

Its purpose is to measure how closely SHAP and LIME agree when identifying important features for a particular prediction.

It uses Spearman's rank correlation coefficient (ρ) to compare the feature rankings generated by both methods.

• High ECI: SHAP and LIME provide similar feature rankings.

• Low ECI: The two methods produce different feature rankings.

The average correlation across the test dataset provides an overall consistency measurement.

The proposed target is an ECI greater than 0.7, indicating strong agreement.

8. Performance Evaluation

The framework evaluates two major aspects:

A. Predictive Performance

• Accuracy

• Precision

• Recall

• F1-score

The paper mentions an expected accuracy range of 90–97% for a Random Forest baseline on the WDBC dataset. This is an expectation, not an experimentally measured result.

B. Explainability Performance

• Explanation Consistency Index (ECI).

• Surrogate decision tree fidelity.

• Agreement between SHAP and LIME feature rankings.

Surrogate fidelity measures how frequently the decision tree produces the same prediction as the original machine learning model.

Important The results section contains illustrative placeholders. The proposed framework has not yet been executed, so no actual experimental results or performance improvements have been demonstrated.

9. Advantages

• Makes black-box machine learning models easier to understand.

• Combines multiple XAI techniques in one framework.

• Provides both global and local explanations.

• Helps identify important features influencing predictions.

• Generates understandable decision rules.

• Introduces ECI to measure agreement between explanation methods.

• Can support transparency, auditing and trust in AI systems.

Conclusion

This paper reviewed the major families of explainable AI techniques and proposed a hybrid framework combining global (SHAP), local (LIME), and rule-based surrogate explanations, unified by a novel Explanation Consistency Index that quantifies agreement between independent explanation methods. Transparency and performance need not be opposing goals — layering complementary explanation techniques over an already-accurate model can deliver both. Future work should extend the framework to deep learning models on image and text data (incorporating Grad-CAM and attention-based methods), validate the ECI against human-expert judgments of explanation quality, and explore real-time explainability dashboards for clinical and financial deployment settings.

References

[1] Ribeiro, M. T., Singh, S., & Guestrin, C. \"Why Should I Trust You?\": Explaining the Predictions of Any Classifier. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2016. doi:10.1145/2939672.2939778 [2] Lundberg, S. M., & Lee, S.-I. A Unified Approach to Interpreting Model Predictions. Advances in Neural Information Processing Systems (NeurIPS), 2017. [3] Selvaraju, R. R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., & Batra, D. Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization. IEEE International Conference on Computer Vision (ICCV), 2017. doi:10.1109/ICCV.2017.74 [4] Tjoa, E., & Guan, C. A Survey on Explainable Artificial Intelligence (XAI): Toward Medical XAI. IEEE Transactions on Neural Networks and Learning Systems, 2021. doi:10.1109/TNNLS.2020.3027314 [5] Bussmann, N., Giudici, P., Marinelli, D., & Papenbrock, J. Explainable AI in Fintech Risk Management. Frontiers in Artificial Intelligence, 2020. doi:10.3389/frai.2020.00026 [6] Rudin, C. Stop Explaining Black Box Machine Learning Models for High Stakes Decisions and Use Interpretable Models Instead. Nature Machine Intelligence, 2019. doi:10.1038/s42256-019-0048-x [7] Doshi-Velez, F., & Kim, B. Towards a Rigorous Science of Interpretable Machine Learning. arXiv:1702.08608, 2017. [8] Adadi, A., & Berrada, M. Peeking Inside the Black-Box: A Survey on Explainable Artificial Intelligence (XAI). IEEE Access, 2018. doi:10.1109/ACCESS.2018.2870052 [9] Arrieta, A. B., et al. Explainable Artificial Intelligence (XAI): Concepts, Taxonomies, Opportunities and Challenges Toward Responsible AI. Information Fusion, 2020. doi:10.1016/j.inffus.2019.12.012 [10] Guidotti, R., Monreale, A., Ruggieri, S., Turini, F., Giannotti, F., & Pedreschi, D. A Survey of Methods for Explaining Black Box Models. ACM Computing Surveys, 2018. doi:10.1145/3236009 [11] Samek, W., Wiegand, T., & Müller, K.-R. Explainable Artificial Intelligence: Understanding, Visualizing and Interpreting Deep Learning Models. arXiv:1708.08296, 2017. [12] Alvarez-Melis, D., & Jaakkola, T. S. On the Robustness of Interpretability Methods. arXiv:1806.08049, 2018. [13] Jain, S., & Wallace, B. C. Attention Is Not Explanation. Proceedings of NAACL-HLT, 2019. [14] Wachter, S., Mittelstadt, B., & Russell, C. Counterfactual Explanations Without Opening the Black Box. Harvard Journal of Law & Technology, 2018. [15] Molnar, C. Interpretable Machine Learning: A Guide for Making Black Box Models Explainable, 2020. [16] Goodman, B., & Flaxman, S. European Union Regulations on Algorithmic Decision-Making and a \"Right to Explanation.\" AI Magazine, 2017. doi:10.1609/aimag.v38i3.2741 [17] Holzinger, A., Langs, G., Denk, H., Zatloukal, K., & Müller, H. Causability and Explainability of Artificial Intelligence in Medicine. WIREs Data Mining and Knowledge Discovery, 2019. doi:10.1002/widm.1312 [18] Miller, T. Explanation in Artificial Intelligence: Insights from the Social Sciences. Artificial Intelligence, 2019. doi:10.1016/j.artint.2018.07.

Copyright

Copyright © 2026 Varshit Khaliya, Sudhanshu Ranjan, Dr. Vishal Shrivastava, Dr. Akhil Pandey. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.