Peramalan Konsentrasi PM2,5 Di Provinsi DKI Jakarta Dengan Pendekatan Model Hybrid Long Short-Term Memory Dan Extreme Gradient Boosting

Duka, Kevin Septian (2026) Peramalan Konsentrasi PM2,5 Di Provinsi DKI Jakarta Dengan Pendekatan Model Hybrid Long Short-Term Memory Dan Extreme Gradient Boosting. Other thesis, INSTITUT TEKNOLOGI SEPULUH NOPEMBER.

Text 5003221019-Undergraduate_Thesis.pdf - Accepted Version Restricted to Registered users only Download (5MB)

Abstract

Particulate Matter berdiameter kurang dari 2,5 μm (PM2,5) merupakan salah satu polutan udara yang paling berbahaya karena dapat menimbulkan berbagai gangguan kesehatan, terutama di wilayah perkotaan dengan aktivitas yang tinggi seperti DKI Jakarta. Peramalan konsentrasi PM2,5 menjadi penting untuk mendukung sistem peringatan dini serta penyusunan kebijakan pengendalian pencemaran udara. Namun, karakteristik data PM2,5 yang bersifat nonlinier, memiliki ketergantungan temporal, serta mengandung missing value menjadi tantangan dalam pemodelannya. Penelitian ini bertujuan untuk meramalkan konsentrasi PM2,5 di Stasiun DKI 1 menggunakan model Hybrid Long Short-Term Memory (LSTM) dan Extreme Gradient Boosting (XGBoost). Data yang digunakan merupakan data harian periode 1 Januari 2022 hingga 31 Oktober 2025 yang terdiri atas konsentrasi PM2,5, polutan pendukung, dan variabel meteorologi. Missing value ditangani menggunakan metode Kalman Smoothing, sedangkan pemilihan fitur dilakukan melalui analisis Mutual Information. Selanjutnya, model LSTM, XGBoost, dan Hybrid LSTM–XGBoost dibangun dengan proses hyperparameter tuning menggunakan Bayesian Optimization. Hasil penelitian menunjukkan bahwa model Hybrid LSTM–XGBoost memberikan performa yang lebih baik dibandingkan model tunggal dengan menghasilkan nilai RMSE sebesar 10,56 dan MAPE sebesar 10,30% pada data pengujian. Meskipun peningkatan akurasi terhadap model pembanding tidak terlalu besar, model hybrid menunjukkan performa yang lebih stabil dalam mengikuti pola perubahan konsentrasi PM2,5. Hasil tersebut menunjukkan bahwa pendekatan hybrid mampu mengombinasikan kemampuan LSTM dalam menangkap ketergantungan temporal dengan kemampuan XGBoost dalam memodelkan hubungan nonlinier, sehingga berpotensi menjadi alternatif yang andal untuk peramalan kualitas udara di DKI Jakarta.
=================================================================================================================================
Particulate Matter with an aerodynamic diameter of less than 2.5 μm (PM2.5) is one of the most hazardous air pollutants due to its significant adverse effects on human health, particularly in densely populated urban areas such as DKI Jakarta. Accurate PM2.5 forecasting is essential for supporting early warning systems and air quality management. However, forecasting PM2.5 remains challenging because of the nonlinear characteristics, temporal dependencies, and missing values commonly found in air quality datasets. This study aims to forecast PM2.5 concentrations at the DKI 1 monitoring station using a hybrid Long Short-Term Memory (LSTM) and Extreme Gradient Boosting (XGBoost) model. Daily air quality and meteorological data collected from January 1, 2022, to October 31, 2025, were used in this study. Missing values were first imputed using the Kalman Smoothing method, while feature selection was performed using Mutual Information analysis. Subsequently, LSTM, XGBoost, and Hybrid LSTM–XGBoost models were developed, with hyperparameter optimization conducted using Bayesian Optimization. The experimental results indicate that the Hybrid LSTM–XGBoost model achieved slightly better forecasting performance, obtaining a testing RMSE of 10.56 and a testing MAPE of 10.30%. Although the improvement over the standalone models was relatively modest, the hybrid model demonstrated more stable performance in capturing fluctuations in PM2.5 concentrations. These findings suggest that combining the temporal learning capability of LSTM with the nonlinear modeling strength of XGBoost provides a reliable approach for PM2.5 forecasting in DKI Jakarta.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Extreme Gradient Boosting, Hybrid LSTM–XGBoost, Kalman Smoothing, Long Short-Term Memory, Peramalan PM2.5.
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines.
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	Kevin Septian Duka Septian Duka
Date Deposited:	04 Aug 2026 02:53
Last Modified:	04 Aug 2026 02:53
URI:	http://repository.its.ac.id/id/eprint/142462

Available Versions of this Item

• Peramalan Konsentrasi PM2,5 Di Provinsi DKI Jakarta Dengan Pendekatan Model Hybrid Long Short-Term Memory Dan Extreme Gradient Boosting. (deposited 31 Jul 2026 06:25)
  • Peramalan Konsentrasi PM2,5 Di Provinsi DKI Jakarta Dengan Pendekatan Model Hybrid Long Short-Term Memory Dan Extreme Gradient Boosting. (deposited 04 Aug 2026 02:53) [Currently Displayed]

Actions (login required)

View Item