Analisis Komparatif Algoritma Machine Learning dengan Pustaka PyTides dalam Peramalan Data Deret Waktu Pasang Surut Air Laut

Noor, Muhammad Hafizh Najwan (2026) Analisis Komparatif Algoritma Machine Learning dengan Pustaka PyTides dalam Peramalan Data Deret Waktu Pasang Surut Air Laut. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221278-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (14MB) | Request a copy

Abstract

Indonesia, sebagai negara kepulauan terbesar di dunia dengan garis pantai yang sangat panjang, menghadapi tantangan krusial dalam pengelolaan wilayah pesisir. Informasi pasang surut air laut di wilayah pesisir memegang peranan vital bagi keselamatan navigasi, aktivitas pelabuhan, hingga mitigasi bencana banjir rob. Namun, kompleksitas dinamika perairan nusantara seringkali sulit diprediksi secara presisi, terutama ketika dihadapkan pada anomali cuaca dan faktor non-linier lainnya. Penelitian ini bertujuan untuk melakukan analisis komparatif antara tiga algoritma machine learning, yaitu Linear Regression, Random Forest, dan XGBoost, dengan metode analisis harmonik yang diterapkan pada pustaka PyTides dalam melakukan peramalan time-series pasang surut air laut di berbagai lokasi (multi-location) di Indonesia. Pengujian dilakukan menggunakan data historis pasang surut dari 10 stasiun pengamatan yang tersebar di seluruh wilayah Indonesia, mencakup data BMKG Maritim dan sensor lapangan Tanjung Sekong. Kinerja setiap model diukur berdasarkan metrik evaluasi kesalahan yang mencakup Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), dan Root Mean Square Error (RMSE) untuk mengidentifikasi metode dengan kinerja terbaik. Penelitian ini berhasil memformulasikan pemrosesan data pasang surut menjadi format deret waktu terstruktur melalui teknik sliding window, rekayasa fitur lag harian, dan encoding konstituen harmonik sintetis (M2, S2, K1, O1, N2), sehingga data oseanografi dapat diadaptasi sebagai input supervised learning yang valid. Baseline PyTides menghasilkan MAE rata-rata 0.093 m pada sembilan stasiun BMKG dengan performa terbaik di Kep. Seribu (MAE 0.023 m) dan terburuk di Probolinggo (MAE 0.429 m), mencerminkan keterbatasan model harmonik murni pada stasiun dengan pengaruh non-astronomis yang kuat. Hasil perbandingan menunjukkan XGBoost sebagai algoritma machine learning terbaik dengan rata-rata MAE 0.070 m pada konfigurasi fitur lag harian dan konstituen harmonik, mampu mengungguli PyTides pada stasiun dengan pengaruh non-astronomis yang kuat seperti Probolinggo (+67.9%) dan Gilimanuk (+36.8%), sementara PyTides tetap superior pada stasiun dengan pola pasang surut reguler yang didominasi komponen astronomis seperti Kep. Seribu dan Stasiun Maritim Ambon dengan selisih MAE hingga 34.2% dan 106.6%.
==================================================================================================================================
Indonesia, as the largest archipelagic country in the world with an extensive coastline, faces crucial challenges in coastal zone management. Tidal information in coastal areas plays a vital role in navigation safety, port operations, and tidal flood disaster mitigation. However, the complexity of water dynamics throughout the archipelago is often difficult to predict accurately, particularly when influenced by weather anomalies and other non-linear factors. This study aims to conduct a comparative analysis of three machine learning algorithms, namely Linear Regression, Random Forest, and XGBoost, as well as the harmonic analysis method implemented in the PyTides library, for forecasting tidal sea-level time-series data at multiple locations across Indonesia. Testing was performed using historical tidal data collected from 10 observation stations distributed throughout Indonesia, including data from BMKG Maritime stations and the Tanjung Sekong field sensor. The performance of each model was evaluated using error metrics, namely Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Root Mean Square Error (RMSE), to identify the best-performing method. This study successfully formulated tidal data processing into a structured time-series format through the application of sliding window techniques, daily lag feature engineering, and synthetic harmonic constituent encoding (M2, S2, K1, O1, and N2), enabling oceanographic data to be adapted as valid supervised learning inputs. The PyTides baseline model produced an average MAE of 0.093 m across nine BMKG stations, achieving its best performance at Kepulauan Seribu (MAE 0.023 m) and its worst performance at Probolinggo (MAE 0.429 m). These results reflect the limitations of a purely harmonic model at stations strongly influenced by non-astronomical factors. Comparative analysis demonstrated that XGBoost is the best-performing machine learning algorithm, achieving an average MAE of 0.070 m under the daily lag and harmonic constituent feature configuration. The model was capable of outperforming PyTides at stations with strong non-astronomical influences, such as Probolinggo (67.9% improvement) and Gilimanuk (36.8% improvement). Nevertheless, PyTides remained superior at stations with regular tidal patterns dominated by astronomical components, such as Kepulauan Seribu and Stasiun Maritim Ambon, with MAE differences reaching 34.2% and 106.6%, respectively. The findings indicate that machine learning approaches, particularly XGBoost, offer significant potential for improving tidal forecasting accuracy in complex coastal environments. At the same time, harmonic analysis methods remain highly effective in locations where tidal behavior is predominantly governed by astronomical forces. Therefore, the selection of forecasting methods should consider the specific oceanographic characteristics of each observation site to achieve optimal prediction performance.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Machine Learning, Pasang Surut Air Laut, Peramalan Time-series, PyTides, Ocean Tides, Time-series Forecasting.
Subjects:	T Technology > T Technology (General) > T174 Technological forecasting T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Muhammad Hafizh Najwan Noor
Date Deposited:	10 Jun 2026 01:12
Last Modified:	10 Jun 2026 01:12
URI:	http://repository.its.ac.id/id/eprint/133616

Actions (login required)

View Item