<![CDATA[This study analyzes and compares the performance of K-Nearest Neighbor (KNN) Regression and Support Vector Regression (SVR) in predicting the reliability of electrical power distribution systems based on the SAIDI and SAIFI indices. The KNN Regression results indicate high sensitivity to the selection of parameter K; the optimal performance was achieved within the range of K = 2–5, while performance degradation occurred at K ≥ 6 due to the loss of locality effect. In contrast, the proposed SVR model (Model SVR), implemented with λ = 0.01 and ε = 0.03 and trained for 5000 epochs, demonstrated more stable and robust performance, achieving a training MAE of 0.126161, a training classification accuracy of 87.5% (7 out of 8 correctly classified samples), and a testing accuracy of 100% (2 out of 2 correctly classified samples). The resulting model coefficients, w0 = −1.038554 and w1 = 0.020590, indicate that SAIDI has a dominant and negative influence on the reliability score, which is physically consistent with the interpretation of outage duration. These findings suggest that, for small-sized datasets, the margin-based SVR approach provides greater robustness and stability compared to the distance-based KNN Regression method, thereby offering a more reliable framework for electrical distribution reliability prediction.]]>
Copyrights © 2026
