Journal of Electronic Science and Technology, Volume. 23, Issue 2, 100315(2025)
Efficient feature selection based on Gower distance for breast cancer diagnosis
Figures & Tables(13)
Fig. 1. Cloud-based feature selection framework using the Gower distance for enhanced breast cancer diagnosis in medical organizations.
Fig. 3. Normalization stage with min-max scaler: (a) before and (b) after min-max normalization.
Fig. 5. Confusion matrices of standard classifiers without feature selection: (a) KNN, (b) NB, (c) SVM, (d) DT, (e) RF, and (f) LR.
Fig. 6. Accuracy comparison of classifiers with and without the proposed feature selection method based on the Gower distance.
Fig. 7. Confusion matrices of standard classifiers with the proposed feature selection method based on the Gower distance: (a) KNN, (b) NB, (c) SVM, (d) DT, (e) RF, and (f) LR.
Fig. 8. Execution time of classifiers with and without the proposed feature selection method.
Table 1. WDBC dataset details.
View tableView in ArticleTable 1. WDBC dataset details.
Parameter Characteristic Feature type Real Dataset characteristics Multivariate Number of cases 569 Number of healthy people 357 Number of unhealthy people 212 Number of features 30 Missing values N\A Classification type Binary classification
Table 2. System environment and setup.
View tableView in ArticleTable 2. System environment and setup.
Specification Description RAM 16 GB Processor 6th Gen Intel® Core™ i7 Development environment Visual Studio Code Software environment Python 3.12
Table 3. Performance evaluation of traditional classifiers without feature selection.
View tableView in ArticleTable 3. Performance evaluation of traditional classifiers without feature selection.
Classifier Metrics Recall (%) Precision (%) F1-score (%) Accuracy (%) KNN 92.10 97.22 94.60 92.98 NB 90.54 93.06 91.78 89.47 DT 89.61 95.83 92.62 90.35 RF 97.10 93.06 95.04 93.86 SVM 89.74 97.22 93.33 91.23 LR 92.00 95.83 93.88 92.10
Table 4. Accuracy results of classifiers and its impact on performance with varying feature ratios and block sizes.
View tableView in ArticleTable 4. Accuracy results of classifiers and its impact on performance with varying feature ratios and block sizes.
Block size Classifier Feature ratio from the dataset 10% 20% 30% 40% 50% 60% 70% 5 samples per block KNN 87.71 89.08 93.22 98.24 90.35 92.10 91.22 NB 85.96 87.08 91.01 96.49 91.22 88.59 90.35 DT 82.45 89.47 92.21 97.36 93.85 94.73 91.22 RF 88.59 92.10 94.73 99.12 94.73 95.61 95.61 SVM 83.34 84.56 92.91 95.61 94.76 92.30 90.74 LR 85.95 88.40 91.50 97.36 90.89 87.89 90.21 10 samples per block KNN 87.71 85.08 91.22 89.47 91.22 92.10 91.22 NB 85.96 85.98 84.21 88.59 89.47 90.33 90.35 DT 80.70 88.59 82.45 92.98 88.59 92.98 93.85 RF 89.47 92.10 92.10 94.73 95.61 95.61 94.73 SVM 84.65 92.40 88.40 93.94 92.20 88.90 86.30 LR 89.20 90.80 92.02 92.98 94.34 94.87 93.33 15 samples per block KNN 88.21 85.08 89.47 87.71 91.22 91.22 92.11 NB 86.96 85.08 85.96 85.08 90.08 90.35 93.21 DT 81.70 88.59 89.47 89.47 92.98 92.98 94.73 RF 88.47 92.98 90.35 92.10 94.73 94.73 95.61 SVM 88.90 90.32 89.02 85.41 89.89 90.87 94.67 LR 90.21 92.56 90.20 91.21 93.43 94.67 96.89 20 samples per block KNN 89.31 85.08 92.22 87.71 93.12 92.10 93.33 NB 86.46 85.99 88.21 85.08 92.47 88.59 90.35 DT 83.40 88.59 89.33 88.59 91.22 92.98 91.22 RF 91.22 92.10 93.85 92.98 94.73 93.85 95.61 SVM 89.20 87.60 89.80 90.65 91.30 87.80 89.50 LR 84.50 90.40 92.31 91.83 92.50 91.12 90.32 25 samples per block KNN 87.71 85.08 89.47 87.71 91.22 92.10 91.22 NB 85.96 85.08 85.96 85.08 89.47 90.11 90.35 DT 83.33 87.71 91.22 88.59 90.35 93.85 92.98 RF 90.35 92.98 92.98 92.10 94.73 95.61 94.73 SVM 84.20 87.60 89.54 89.60 90.30 90.59 90.74 LR 88.80 91.50 90.89 91.23 93.98 94.43 93.20
Table 5. Performance comparison with recently related works on the WDBC dataset.
View tableView in ArticleTable 5. Performance comparison with recently related works on the WDBC dataset.
Year Reference Feature selection Number of selected features Classifier Accuracy (%) Execution time 2022 [ 26 ]Univariate and recursive 16 Deep extreme gradient descent optimization 98.73 N/A 2023 [ 17 ]PCC 15 KNN 91.2 N/A RF 96.5 LR 94.7 XGBoost 97.4 2023 [ 20 ]PCA 16 SVM 98.07 N/A DT 94.20 KNN 96.84 RF 94.20 MLP 97.54 NB 91.04 LR 98.42 LR+SVM 98.77 2023 [ 27 ]Chi-square 15 MLP 95 N/A LR 92 KNN 96 SVM 92 RF 94 2023 [ 28 ]Gorilla troops optimization (GTO) 30 Deep Q learning (DQL) 98.88 N/A 2023 [ 29 ]ESO 12 RF 98.95 4 s 2024 [ 30 ]EHO 18 KNN 97.96 N/A 2024 Proposed in this paper Gower distance 12 KNN 98.24 0.1472 msNB 96.49 0.0087 msDT 97.36 0.0021 msRF 99.12 0.0583 msSVM 95.61 0.0262 msLR 97.36 0.0018 ms
Tools
Get Citation
Copy Citation Text
Salwa Shakir Baawi, Mustafa Noaman Kadhim, Dhiah Al-Shammary. Efficient feature selection based on Gower distance for breast cancer diagnosis[J]. Journal of Electronic Science and Technology, 2025, 23(2): 100315
Paper Information
Category:
Received: Dec. 3, 2024
Accepted: Apr. 21, 2025
Published Online: Jun. 16, 2025
The Author Email: Mustafa Noaman Kadhim (mustafa.noaman@qu.edu.iq)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20