<![CDATA[The accuracy of project duration forecasting methods varies across different phases of project execution. This study evaluates the temporal accuracy dynamics of three Earned Value Management-based methods—Earned Value Method (EVM), Earned Schedule Method (ESM), and Kalman Earned Value Method (KEVM)—in predicting the final duration of small-scale building construction projects. Data from two school building construction projects in the Province of Central Kalimantan, Indonesia, each with a planned and actual duration of 20 weeks, were used as case studies. Accuracy was measured using Mean Absolute Percentage Error (MAPE) and evaluated based on Lewis (1982) categories. Analysis was conducted for the full project duration and separately for two phases divided at the 50% BPD threshold. Results show that ESM consistently produced the highest overall accuracy (MAPE 2.95% and 3.77%, Very Good category), while KEVM—using Vaudrey’s (2008) simplified constant C—showed the lowest accuracy in Phase 1 but achieved the best accuracy in Phase 2 (MAPE 2.42% and 3.36%), surpassing both EVM and ESM. This finding demonstrates that the Kalman Filter algorithm becomes increasingly effective as actual data accumulates over project execution, making KEVM particularly suitable for project monitoring in the latter phases of construction. These findings offer practical value for construction stakeholders seeking to strengthen the reliability of duration forecasting, supporting the delivery of resilient infrastructure for school building projects that is more efficient and timely. Abstrak Akurasi metode prediksi durasi Proyek Bervariasi sepanjang fase pelaksanaan. Penelitian ini mengevaluasi dinamika akurasi temporal tiga metode berbasis Earned Value Management (EVM), Earned Schedule Method (ESM), dan Kalman Earned Value Method (KEVM) dalam memprediksi durasi akhir proyek konstruksi bangunan gedung skala kecil. Data dari dua Proyek Bangunan gedung sekolah di Provinsi Kalimantan Tengah, dengan durasi rencana dan realisasi masing-masing 20 minggu, digunakan sebagai studi kasus. Akurasi diukur menggunakan Mean Absolute Percentage Error (MAPE) dengan kategori penilaian Lewis (1982). Analisis dilakukan untuk keseluruhan durasi proyek dan dibagi menjadi dua tahap berdasarkan batas 50% BPD. Hasil menunjukkan bahwa ESM secara konsisten menghasilkan akurasi tertinggi secara keseluruhan (MAPE 2,95% dan 3,77%, kategori Sangat Baik), sedangkan KEVM—menggunakan konstanta C hasil penyederhanaan Vaudrey (2008)—menunjukkan akurasi terendah pada Tahap 1, namun mencapai akurasi terbaik pada Tahap 2 (MAPE 2,42% dan 3,36%), melampaui EVM dan ESM. Temuan ini mengindikasikan bahwa algoritma Kalman Filter semakin efektif seiring bertambahnya data aktual, sehingga KEVM sangat sesuai untuk pemantauan pada fase akhir proyek konstruksi. Temuan ini memberikan nilai praktis bagi pemangku kepentingan konstruksi dalam meningkatkan keandalan prediksi durasi proyek, guna mendukung penyelenggaraan resilient infrastructure bangunan gedung sekolah yang lebih efisien dan tepat waktu.]]>
