Peningkatan jumlah kunjungan pasien di RSUD Syarifah Ambami Rato Ebu Bangkalan menyebabkan antrean panjang pada layanan pendaftaran rawat jalan yang memicu fenomena abandonment, yaitu pasien meninggalkan antrean sebelum memperoleh pelayanan. Penelitian ini bertujuan mengidentifikasi faktor operasional dan perilaku yang memengaruhi probabilitas abandonment serta membangun model probabilitas menggunakan Logistic Regression (LR) dan Gradient Boosted Trees (GBT) pada sistem antrean single queue. Penelitian menggunakan pendekatan observasional kuantitatif terhadap 1.246 observasi pasien dengan variabel headcount depan, laju selesai pelayanan, proksi waktu tunggu (Proksi waktu tunggu), kategori hari, dan jam pelayanan. Hasil penelitian menunjukkan bahwa variabel laju selesai 5 menit dan jam pelayanan 11.00 berpengaruh signifikan terhadap probabilitas abandonment. Model LR setelah kalibrasi menjadi model terbaik dengan accuracy 75,10%, AUC 0,664, LogLoss 0,439, dan Brier Score 0,086, sedangkan GBT setelah kalibrasi memperoleh accuracy 82,33%, AUC 0,657, LogLoss 0,533, dan Brier Score 0,100. Meskipun GBT memiliki accuracy lebih tinggi, model LR kalibrasi dipilih sebagai model terbaik karena menghasilkan kualitas probabilitas yang lebih stabil dan mendekati kondisi aktual berdasarkan nilai AUC, LogLoss, dan Brier Score. Hasil tersebut menunjukkan bahwa kalibrasi probabilitas mampu meningkatkan kualitas probabilitas prediksi sehingga model lebih layak digunakan sebagai dasar pengambilan keputusan operasional rumah sakit.


Deskripsi Lain

The increasing number of patient visits at Syarifah Ambami Rato Ebu Bangkalan Regional General Hospital has caused long queues at outpatient registration services, triggering the abandonment phenomenon, where patients leave the queue before receiving service. This study aims to identify operational and behavioral factors that influence the probability of abandonment and build a probability model using Logistic Regression (LR) and Gradient Boosted Trees (GBT) in a single queue system. The study used a quantitative observational approach to 1,246 patient observations with variables of front headcount, service completion rate, waiting time proxy, day category, and service hours. The results showed that the variables of 5-minute completion rate and 11:00 service hours significantly influenced the probability of abandonment. The LR model after calibration was the best model with an accuracy of 75.10%, an AUC of 0.664, a LogLoss of 0.439, and a Brier Score of 0.086, while the GBT model after calibration achieved an accuracy of 82.33%, an AUC of 0.657, a LogLoss of 0.533, and a Brier Score of 0.100. Although the GBT model had higher accuracy, the calibrated LR model was chosen the best model because it produced more stable probability quality and was closer to actual conditions based on AUC, LogLoss, and Brier Score values. These results indicate that probability calibration can improve the quality of predicted probability, making the model more suitable for use a basis for hospital operational decision-making.

• PEMODELAN DAN KALIBRASI PROBABILITAS ABANDONMENT PASIEN BERBASIS PROKSI WAKTU TUNGGU PADA SISTEM ANTREAN SINGLE QUEUE MENGGUNAKAN MODEL LOGISTIC REGRESSION DAN GRADIENT BOOSTED TREES (Studi Kasus: RSUD Syarifah Ambami Rato Ebu, Bangkalan)