Scholarlink

<p>Breast cancer (BC) is one of the most prevalent cancers among women. Early detection of this malignancy is critical for guiding effective treatment. Currently various machine learning (ML) classification models have been employed for BC detection using patient health records. However the effectiveness of each model depends on multiple factors including model configurations parameter settings and input attribute types. To overcome the limitations of individual models and the shortcomings of existing ML-based BC detection approaches this research proposes a SHapley Additive exPlanations (SHAP) and Bidirectional Long Short-Term Memory (BiLSTM) model enhanced with a scaled dot-product attention mechanism for BC detection. In this research Principal Component Analysis (PCA) was employed to establish a new orthogonal basis space where each axis represents a principal component. The incorporation of SHAP analysis in BC detection augments the interpretability and validity of ML models. It facilitates the critical evaluation of model behavior ensuring that predictions are not only accurate but also scientifically and clinically justifiable. The BiLSTM model dynamically prioritizes critical components within the BC data that are more likely to contain malignant patterns. Additionally we construct a voting ensemble model that utilizes the average of probabilities to enhance the diagnostic accuracy of AdaBoost Bagging random forest gradient boosting machine XGBoost stacking and rotation forest. To evaluate the performance of the proposed model we conducted experiments using the Wisconsin Breast Cancer data. The results indicate that compared to nine other ensemble models the Extra Trees model achieves the highest accuracy for BC detection. Specifically when compared to the best-performing single models C5.0 Decision Tree Artificial Neural Network and J48 the proposed ensemble Extra Trees model improves accuracy by 3.54% 1.62% and 2.91% respectively. Furthermore when benchmarked against state-of-the-art techniques our proposed model achieved an accuracy of 99.59% surpassing the existing method which achieved 95.52%. The integration of PCA attention-based BiLSTM and the Extra Trees ensemble reduces model development time. Specifically our Extra Trees classifier requires only 0.069&#xa0;s to train on the WisconsinBreast Cancer dataset, considerably less than alternative ensembles such as Rotation Forest of 0 .420&#xa0;s and XGBoost of 7.83&#xa0;s. This efficiency arises because Extra Trees randomly select split thresholds rather than exhaustively searching for optimal splits, thereby reducing computational complexity. The analysis underscores “Uniformity of cell size” and “Uniformity of cell shape” as the most pivotal features influencing the model’s outcomes. These findings corroborate established pathological knowledge, wherein deviations in cell size and shape serve as hallmark indicators of BC malignancy. The proposed model offers a highly reliable, robust, and safer alternative for BC detection, demonstrating significant improvements over current diagnostic procedures.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhanced breast cancer detection via Shapley additive explanations and attention-based bidirectional LSTM with ensemble learning models

  • 期刊论文

  • 发表日期：

    2026年3月7日

查看全文
  • Yakub Kayode Saheed,
  • Mokhairi Makhtar

摘要

Breast cancer (BC) is one of the most prevalent cancers among women. Early detection of this malignancy is critical for guiding effective treatment. Currently various machine learning (ML) classification models have been employed for BC detection using patient health records. However the effectiveness of each model depends on multiple factors including model configurations parameter settings and input attribute types. To overcome the limitations of individual models and the shortcomings of existing ML-based BC detection approaches this research proposes a SHapley Additive exPlanations (SHAP) and Bidirectional Long Short-Term Memory (BiLSTM) model enhanced with a scaled dot-product attention mechanism for BC detection. In this research Principal Component Analysis (PCA) was employed to establish a new orthogonal basis space where each axis represents a principal component. The incorporation of SHAP analysis in BC detection augments the interpretability and validity of ML models. It facilitates the critical evaluation of model behavior ensuring that predictions are not only accurate but also scientifically and clinically justifiable. The BiLSTM model dynamically prioritizes critical components within the BC data that are more likely to contain malignant patterns. Additionally we construct a voting ensemble model that utilizes the average of probabilities to enhance the diagnostic accuracy of AdaBoost Bagging random forest gradient boosting machine XGBoost stacking and rotation forest. To evaluate the performance of the proposed model we conducted experiments using the Wisconsin Breast Cancer data. The results indicate that compared to nine other ensemble models the Extra Trees model achieves the highest accuracy for BC detection. Specifically when compared to the best-performing single models C5.0 Decision Tree Artificial Neural Network and J48 the proposed ensemble Extra Trees model improves accuracy by 3.54% 1.62% and 2.91% respectively. Furthermore when benchmarked against state-of-the-art techniques our proposed model achieved an accuracy of 99.59% surpassing the existing method which achieved 95.52%. The integration of PCA attention-based BiLSTM and the Extra Trees ensemble reduces model development time. Specifically our Extra Trees classifier requires only 0.069 s to train on the WisconsinBreast Cancer dataset, considerably less than alternative ensembles such as Rotation Forest of 0 .420 s and XGBoost of 7.83 s. This efficiency arises because Extra Trees randomly select split thresholds rather than exhaustively searching for optimal splits, thereby reducing computational complexity. The analysis underscores “Uniformity of cell size” and “Uniformity of cell shape” as the most pivotal features influencing the model’s outcomes. These findings corroborate established pathological knowledge, wherein deviations in cell size and shape serve as hallmark indicators of BC malignancy. The proposed model offers a highly reliable, robust, and safer alternative for BC detection, demonstrating significant improvements over current diagnostic procedures.