Pengembangan Model Prediksi Durasi Pekerjaan Civil Works Pada Proyek SUTT 150 kV Dengan Pendekatan Multiple Linear Regression, Random Forest, Dan XGBoost You’re out of attachments for now. Start a free Plus trial for more, or try again tomorrow after 2:13?PM. Try Plus free
Mohammad Irsyad Husain, Ir. Tantri Nastiti Handayani, S.T., M.Eng., Ph.D., IPM. ; Ir. Akhmad Aminullah, S.T., M.T., Ph.D., IPU.
2026 | Tesis | S2 Teknik Sipil
Pekerjaan civil works pada konstruksi tower transmisi listrik merupakan tahapan penting yang menentukan kesiapan pekerjaan stringing. Keterlambatan penyelesaian pekerjaan ini dapat berdampak pada mundurnya jadwal konstruksi, terganggunya aliran kas kontraktor, keterlambatan penyelesaian proyek, serta potensi kerugian bagi pemilik proyek akibat belum beroperasinya sistem transmisi sesuai target. Maka diperlukan model prediksi durasi pekerjaan civil works yang mampu memberikan estimasi waktu secara lebih terukur berdasarkan karakteristik teknis, kondisi lapangan, sumber daya, dan logistik proyek. Penelitian ini bertujuan untuk membangun dan membandingkan model prediksi durasi pekerjaan civil works konstruksi tower menggunakan metode Multiple Linear Regression (MLR), Random Forest, dan Extreme Gradient Boosting (XGBoost).
Objek penelitian adalah proyek SUTT 150 kV Interkoneksi Kaltim - Kaltimra dengan jumlah data sebanyak 312 titik tower. Variabel dependen yang digunakan adalah durasi pekerjaan civil works, yaitu durasi dari pekerjaan pondasi dimulai sampai tower berdiri. Variabel independen mencakup aspek struktur tower dan pondasi, kondisi medan, prioritas pekerjaan, sumber daya, serta logistik material stub dan tower. Data dibagi menjadi data pelatihan 80?n data pengujian 20%. Evaluasi performa model dilakukan menggunakan indikator MAE, RMSE, MAPE, dan koefisien determinasi R².
Hasil penelitian menunjukkan model MLR M3R menghasilkan nilai R² sebesar 0,541 pada data training dan 0,611 pada data testing, dengan MAE, RMSE, dan MAPE pengujian masing-masing sebesar 59,961 hari, 84,537 hari, dan 33,031%. Model Random Forest baseline menghasilkan R² pengujian sebesar 0,646, MAE sebesar 54,305 hari, RMSE sebesar 80,647 hari, dan MAPE sebesar 27,844%, tetapi menunjukkan kecenderungan overfitting karena terdapat selisih R² yang cukup besar antara data training dan testing. Model XGBoost baseline menghasilkan R² pengujian tertinggi sebesar 0,666 dan RMSE terendah sebesar 78,311 hari, sedangkan XGBoost tuned menghasilkan MAE dan MAPE terendah, masing-masing sebesar 51,999 hari dan 24,703%. Berdasarkan keseimbangan antara akurasi dan kemampuan generalisasi, XGBoost baseline ditetapkan sebagai model prediksi terbaik secara keseluruhan, sedangkan XGBoost tuned lebih sesuai apabila pengurangan kesalahan rata-rata menjadi prioritas. MLR M3R tetap memiliki keunggulan dalam stabilitas dan interpretasi hubungan antarvariabel. Hasil penelitian ini dapat digunakan sebagai dasar pengembangan alat bantu berbasis data dalam perencanaan, pengendalian, dan evaluasi proyek transmisi listrik, khususnya untuk memperkirakan durasi pekerjaan civil works sebelum pekerjaan stringing dilaksanakan.
Civil works in
transmission tower construction represent a critical stage that determines the
readiness of subsequent stringing activities. Delays in completing civil works
may affect the overall construction schedule, disrupt contractors’ cash flow,
extend project completion time, and potentially cause losses to the project
owner due to delayed operation of the transmission system. Therefore, a
duration prediction model is required to provide a more measurable estimation
of civil works completion time based on technical characteristics, site
conditions, resources, and project logistics. This study aims to develop and
compare civil works duration prediction models for transmission tower
construction using Multiple Linear Regression (MLR), Random Forest, and Extreme
Gradient Boosting (XGBoost) methods.
The research
object is the 150 kV Kaltim - Kaltimra Interconnection transmission line
project, consisting of 312 tower point data. The dependent variable is the
duration of civil works, defined as the period from the start of foundation
work until the tower is erected. The independent variables include tower and
foundation structural characteristics, terrain conditions, work priority,
resources, and stub and tower material logistics. The dataset was divided into
80% training data and 20% testing data. Model performance was evaluated using
MAE, RMSE, MAPE, and the coefficient of determination R².
The results show that the MLR M3R model achieved R² values of 0.541 on the training data and 0.611 on the testing data, with testing MAE, RMSE, and MAPE values of 59.961 days, 84.537 days, and 33.031%, respectively. The baseline Random Forest model achieved a testing R² of 0.646, an MAE of 54.305 days, an RMSE of 80.647 days, and a MAPE of 27.844%; however, it indicated a tendency toward overfitting because of the relatively large difference between its training and testing R² values. The baseline XGBoost model achieved the highest testing R² of 0.666 and the lowest RMSE of 78.311 days, whereas the tuned XGBoost model produced the lowest MAE and MAPE values of 51.999 days and 24.703%, respectively. Based on the balance between predictive accuracy and generalization capability, the baseline XGBoost model was selected as the best overall prediction model, while the tuned XGBoost model is more suitable when minimizing average prediction errors is prioritized. Meanwhile, the MLR M3R model remains advantageous in terms of stability and interpretability of the relationships among variables. These findings provide a foundation for developing data-driven decision-support tools for the planning, control, and evaluation of transmission line construction projects, particularly for estimating civil works duration before stringing activities are carried out.
Kata Kunci : durasi civil works, konstruksi tower transmisi, Multiple Linear Regression, Random Forest, XGBoost, prediksi durasi proyek
