<![CDATA[Operation Search and Rescue (SAR) in phase DETRESFA demands fast and accurate decision-making because it involves real, life-threatening situations. The Medan Basarnas UPT faces challenges in classifying four main categories of incidents: ship accidents (Y1), accidents requiring special handling (Y2), natural disasters (Y3), and conditions endangering humans (Y4), which have so far been done manually and subjectively. This study aims to build a decision support system based on data collection. machine learning to improve the efficiency of resource deployment through objective classification of emergency conditions. Performance comparisons were conducted between the algorithms Random ForestAnd Support Vector Machine(SVM) based on operational features such asresponse time, number of victims, number of personnel, and distance of the incident. The test results show thatRandom Forestprovides superior performance compared to SVM across all evaluation metrics, with accuracy 86.4%, AUC value 93.9%, F1-score 85.5%, And Matthews Correlation Coefficient (MCC) 0.759. AnalysisConfusion Matrixconfirm that Random Foresthas better stability in recognizing operational feature patterns for most target categories, including its more consistent ability in handling less dominant classes than SVM. Although the Y2 category is still a challenge for both models, Random Forestproven to be much more robust with an accuracy of 49.6% compared to SVM which only achieved 16.5%. This research proves that Random Forestis a more reliable and consistent model to support SAR practitioners in improving the accuracy of field responses, efficiency of resource deployment, and minimizing the risk of loss of life.]]>
Copyrights © 2026
