<![CDATA[Abstrak Penelitian ini mengembangkan model machine learning untuk memprediksi mortalitas harian (jumlah kematian) ayam broiler 1–7 hari ke depan secara multi-horizon (H+1–H+7) menggunakan data 12 kandang di Sukabumi selama lima siklus produksi (Juli 2024–Juli 2025). Data dipraproses melalui imputasi nilai hilang (Random Forest), penanganan outlier (IQR dan winsorizing), normalisasi Z-score, serta seleksi fitur (Pearson dan ReliefF). Support Vector Regression (SVR) dan Decision Tree Regression (DTR) dibandingkan dengan hasil menunjukkan SVR unggul untuk prediksi jangka pendek (H+1–H+2; R² = 0,842–0,760), tetapi performanya menurun pada horizon yang lebih panjang. Sebaliknya, DTR lebih stabil pada horizon menengah–panjang (H+5–H+7; R² ≈ 0,683–0,696). Faktor dominan yang berkaitan dengan mortalitas adalah umur dan bobot rata-rata, serta kondisi kandang seperti ventilasi/kecepatan angin, kepadatan tebar, NH₃, dan suhu. Evaluasi dilakukan dengan repeated holdout 70/30 (10 repetisi) dan 5-fold cross-validation pada data latih, mendukung prototipe sebagai peringatan dini. Kata kunci: ayam broiler, decision tree regression, mortalitas, multi-horizon forecasting, support vector regression Abstract This study develops a machine-learning model to predict daily broiler mortality (death counts) 1–7 days ahead using a multi-horizon approach (H+1–H+7), based on data from 12 broiler houses in Sukabumi across five production cycles (July 2024–July 2025). Data were preprocessed using missing-value imputation (Random Forest), outlier handling (IQR and winsorizing), Z-score normalization, and feature selection (Pearson correlation and ReliefF). Support Vector Regression (SVR) and Decision Tree Regression (DTR) were compared. Results show that SVR outperformed DTR for short-term prediction (H+1–H+2; R² = 0.843–0.760), but its performance declined at longer horizons. In contrast, DTR was more stable for medium-to-long horizons (H+5–H+7; R² ≈ 0.683–0.696). Dominant factors associated with mortality included age and average body weight, as well as housing conditions such as ventilation/wind speed, stocking density, NH₃, and temperature. Evaluation used repeated 70/30 holdout (10 repetitions) and 5-fold cross-validation on the training data, supporting a prototype as an early warning tool. Keywords: broiler chicken, decision tree regression, mortality, multi-horizon forecasting, support vector regression]]>
