Leveraging Local Interpretable Model-Agnostic Explanations for Explainable AI in Healthcare : Improving Breast Cancer Prediction Models

Abstract

Early diagnosis and effective treatment of breast cancer are critical factors in improving survival rates among patients. Consequently, developing accurate predictive models for breast cancer occurrence has become increasingly important in healthcare. This study employs advanced machine learning techniques to analyze a comprehensive dataset comprising 5025 instances, which includes 16 variables such as demographic information, tumor size, hormone receptor status, and survival time. These variables provide a rich foundation for building a predictive model aimed at assessing the likelihood of breast cancer development in patients.

To construct this predictive model, we utilize the Random Forest algorithm, which is widely recognized for its robustness and high accuracy in handling complex datasets. Random Forest operates by creating multiple decision trees and aggregating their predictions, which generally leads to improved performance over single-tree models. However, despite its predictive strength, one notable limitation of the Random Forest algorithm is its lack of interpretability. This presents challenges for healthcare professionals, as understanding the underlying reasons behind specific predictions is crucial for informed clinical decision-making.

To overcome this interpretability challenge, we propose an integrated approach that combines the Random Forest model with Local Interpretable Model-Agnostic Explanations (LIME). LIME is designed to provide local explanations for individual predictions, thereby enhancing the transparency of the model. By generating insights into how specific variables influence predictions for individual cases, LIME enables healthcare professionals to better understand the model’s decision-making process.

Our proposed methodology not only preserves the high accuracy achieved by the Random Forest model but also enriches it with valuable interpretative insights. This dual benefit allows healthcare professionals to place greater trust in the model and utilize its predictions more effectively in clinical settings. Ultimately, our approach bridges the gap between sophisticated machine learning predictions and practical applications in healthcare, paving the way for improved patient outcomes through informed decision-making.