<![CDATA[Breast cancer is one of the most prevalent types of cancer in Indonesia, making accurate early detection highly crucial to reduce mortality rates. However, classification effectiveness is often hindered by high-dimensional data and class imbalance issues, which can introduce bias into predictive models. This study proposes the integration of the Decision Tree algorithm with hybrid sampling techniques and Particle Swarm Optimization (PSO). PSO is employed to perform a dual role, namely simultaneous feature selection and hyperparameter tuning. Experimental results show that using 30 particles in PSO successfully reduced the data dimensionality significantly from 30 features to only 6 essential features: texture1, symmetry1, radius2, area3, smoothness3, and symmetry3. The combination of SMOTE-Tomek and PSO-based feature selection emerged as the best-performing scenario, achieving an accuracy of 99.68% while producing only one False Negative prediction. In addition to superior precision, the model demonstrated remarkable computational efficiency with an average latency of 0.0036 ms and a throughput of 274,897 samples per second. Explainable AI (XAI) analysis using SHAP confirmed that area3 was the most dominant feature, which is clinically consistent with indicators of cancer cell proliferation. This study proves that the synergy between data balancing techniques and metaheuristic optimization can produce an accurate, transparent, and efficient diagnostic model suitable for real-time medical implementation.]]>
Copyrights © 2026
