Pengaruh Hiperparameter dan Variabel Eksogen pada Prediksi Multi-Langkah Kecepatan Angin menggunakan LightGBM
DOI:
https://doi.org/10.14421/fourier.2025.151.1-16Keywords:
horizon, hyperparameter configuration, Direct approach, Recursive approachAbstract
Penelitian ini menganalisis pengaruh konfigurasi hiperparameter dan variabel eksogen terhadap prediksi kecepatan angin multi-langkah menggunakan algoritma LightGBM dengan strategi Recursive dan Direct. Hasil penelitian menunjukkan bahwa panjang horizon prediksi merupakan salah satu faktor utama yang mempengaruhi tingkat kesalahan prediksi, dimana nilai kesalahan meningkat secara konsisten dari horizon 1 langkah hingga 30 langkah. Penambahan variabel eksogen pada model berupa suhu minimum dan kelembaban relatif terbukti mampu meningkatkan kinerja model pada seluruh horizon, dengan dampak yang lebih besar pada horizon menengah dan panjang. Selain itu, tuning hiperparameter menunjukkan pengaruh yang bergantung pada horizon, dimana tuning memberikan manfaat yang lebih jelas pada prediksi jangka panjang. Secara keseluruhan, hasil penelitian ini menunjukkan bahwa kinerja prediksi kecepatan angin tidak hanya dipengaruhi oleh pemilihan algoritma, tetapi juga dipengaruhi oleh kombinasi strategi prediksi, variabel eksogen, konfigurasi hiperparameter, dan panjang horizon prediksi.
Kata Kunci: horizon, konfigurasi hiperparameter, pendekatan Direct, pendekatan Recursive.
Abstract
This study analyzes the effect of hyperparameter configuration and exogenous variables on multi-step wind speed prediction using the LightGBM algorithm with Recursive and Direct strategies. The results show that the length of the prediction horizon is one of the main factors affecting the level of prediction error, with the error increasing consistently from 1 step to 30 steps. The addition of exogenous variables to the model, such as minimum temperature and relative humidity, has been shown to improve model performance across all horizons, with a greater impact on medium- and long-horizon forecasts. In addition, hyperparameter tuning exhibits a horizon-dependent effect, with greater benefits for long-term predictions. Overall, the results of this study indicate that wind speed prediction performance is influenced not only by the choice of algorithm but also by the combination of prediction strategy, exogenous variables, hyperparameter configuration, and prediction horizon length.
Keywords: horizon, kyperparameter configuration, Direct approach, Recursive approach.
Downloads
References
N. Fitriyati, Prediksi Kecepatan Angin Menggunakan Filter Kalman. Tangerang Selatan: UIN Jakarta Press, 2024.
F. Rahman, T. E. Sutanto, and N. Fitriyati, “Web Traffic Anomaly Detection using Stacked Long Short-Term Memory,” Inpr. Indones. J. Pure Appl. Math., vol. 3, no. 2, pp. 112–121, 2021, https://doi.org/10.15408/inprime.v3i2.21879.
R. A. Fahrezi, M. Y. Wijaya, and N. Fitriyati, “Prediksi Harga Penutupan Saham Bank Central Asia: Implementasi Algoritma Long Short-Term Memory Dan Perbandingannya Dengan Support Vector Machine,” J. Lebesgue J. Ilm. Pendidik. Mat. Mat. dan Stat., vol. 5, no. 1, pp. 452–464, 2024, https://doi.org/10.46306/lb.v5i1.582.
S. S. Band et al., “A two-stage deep learning-based hybrid model for daily wind speed forecasting,” Heliyon, vol. 11, no. 1, Jan. 2025, https://doi.org/10.1016/j.heliyon.2024.e41026.
M. Paula, C. Marilaine, F. José Nuno, and C. Wallace, “Predicting Long-Term Wind Speed in Wind Farms of Northeast Brazil: A Comparative Analysis Through Machine Learning Models,” IEEE Lat. Am. Trans., vol. 18, no. 11, pp. 2011–2018, 2020, https://doi.org/10.1109/TLA.2020.9398643.
J. Marc et al., “Optimizing Wind Power Forecasting Using Machine Learning: A Comparative Study with Emphasis on LightGBM for Predictive Maintenance,” Int. J. Adv. Eng. Res. Sci., vol. 12, no. 6, pp. 69–80, 2025, https://doi.org/10.22161/ijaers.126.7.
J. Park and E. A. Hwang, “A Two-Stage Multistep-Ahead Electricity Load Forecasting Scheme Based on LightGBM and Attention-BiLSTM,” Sensors, vol. 21, p. 7697, 2021, https://doi.org/10.3390/s21227697.
Y. Shi, G. Ke, Z. Chen, S. Zheng, and T.-Y. Liu, “Quantized Training of Gradient Boosting Decision Trees,” in 36th Conference on Neural Information Processing Systems (NeurIPS 2022), 2023.
G. Ke et al., “LightGBM: A Highly Efficient Gradient Boosting Decision Tree,” in 31st Conference on Neural Information Processing Systems (NIPS 2017), 2017, pp. 1–9.
I. U. Siloko and U. O. O., “A statistical study of wind speed and its connectivity with relative humidity and temperature in Ughelli, Delta State, Nigeria,” Sci. World J., vol. 18, no. 3, pp. 404–413, 2023, https://doi.org/10.4314/swj.v18i3.13.
S. Ben Taieb, G. Bontempi, A. F. Atiya, and A. Sorjamaa, “A review and comparison of strategies for multi-step ahead time series forecasting based on the NN5 forecasting competition,” Expert Syst. Appl., vol. 39, no. 8, pp. 7067–7083, 2012, https://doi.org/10.1016/j.eswa.2012.01.039.
S. Ben Taieb and R. J. Hyndman, “Recursive and direct multi-step forecasting: the best of both worlds,” [Nama Jurnal/Konferensi], 2012.
T. Mcelroy, “When are Direct Multi-step and Iterative Forecasts Identical?,” J. Forecast., vol. 34, pp. 315–336, 2015, https://doi.org/10.1002/for.2321.
S. Ben Taieb and G. Bontempi, “Recursive multi-step time series forecasting by perturbing data,” in 11th IEEE International Conference on Data Mining, ICDM 2011, Vancouver, BC, Canada, December 11-14, 2011, pp. 695–704, https://doi.org/10.1109/ICDM.2011.123.
A. Lazcano, J. E. Sandubete, and M. A. Jaramillo-Morán, “A comparative framework for multi-horizon time series forecasting: Neural networks with adaptive preprocessing,” Mach. Learn. with Appl., vol. 22, p. 100781, 2025, https://doi.org/10.1016/j.mlwa.2025.100781.
J. Gong, Z. Qu, Z. Zhu, H. Xu, and Q. Yang, “Ensemble models of TCN-LSTM-LightGBM based on ensemble learning methods for short-term electrical load forecasting,” Energy, vol. 318, p. 134757, 2025, https://doi.org/10.1016/j.energy.2025.134757.
J. A. Ilemobayo et al., “Hyperparameter Tuning in Machine Learning: A Comprehensive Review,” J. Eng. Res. Reports, vol. 26, no. 6, pp. 388–395, 2024, https://doi.org/10.9734/jerr/2024/v26i61188.
R. J. Hyndman and G. Athanasopoulos, Forecasting: Principles and Practice (3rd ed.). 2021. [Online]. Available: https://otexts.com/fpp3/
R. Green, H. Day, Z. A. Abdallah, and T. M. S. Filho, “Epistemic Error Decomposition for Multi-step Time Series Forecasting: Rethinking Bias – Variance in Recursive and Direct Strategies,” arXiv, vol. 2511.11461, 2025.
Y. Zhang, J. Wang, and X. Wang, “Review on probabilistic forecasting of wind power generation,” Renew. Sustain. Energy Rev., vol. 32, pp. 255–270, 2014, https://doi.org/10.1016/j.rser.2014.01.033.
H. Liu, H.-Q. Tian, C. Chen, and Y. Li, “A hybrid statistical method to predict wind speed and wind power,” Renew. Energy, vol. 35, no. 8, pp. 1857–1861, 2010, https://doi.org/10.1016/j.renene.2009.12.011.
P. Probst, A.-L. Boulesteix, and B. Bischl, “Tunability: Importance of Hyperparameters of Machine Learning Algorithms,” J. Mach. Learn. Res., vol. 20, pp. 1–32, 2019.
C. Zhang, S. Bengio, M. Hardt, B. Recht, and O. Vinyals, “Understanding deep learning (still) requires rethinking generalization,” Commun. ACM, vol. 64, no. 3, pp. 107–115, Feb. 2021, https://doi.org/10.1145/3446776.
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2026 Nina Fitriyati, Muhaza Liebenlito, Salma Hasna Tsabitah
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
