KONVERGENSI

Arsitektur U-Net Untuk Mendeteksi Titik Api Kebakaran Hutan Dan Lahan Di Kalimantan Tengah Menggunakan Satelit Himawari 8

2025-02-05T07:34:46+00:00

ABSTRACT

Early detection of fire hotspots is crucial to prevent the spread of fires and to mitigate the resulting environmental impacts. This study proposes the use of the U-Net model for detecting fire hotspots on Himawari 8 satellite imagery, specifically utilizing Band 7, which is sensitive to thermal radiation. Fire hotspot data from MODIS and VIIRS were employed as the ground truth to validate the model predictions. The U-Net model was trained on satellite imagery with binary crossentropy as the loss function and was evaluated using Area Under Curve (AUC), Percentage Correct (PC), Omission Error (OE), and Commission Error (CE) as evaluation metrics. The training results demonstrate that the model is capable of detecting fire hotspots with high accuracy, achieving an AUC of over 0.85 on the validation data after 300 epochs. Moreover, PC increased and OE decreased, indicating an improvement in the model's ability to accurately detect fire hotspots. In conclusion, the U-Net model developed in this study has shown good performance in detecting fire hotspots.

Keywords: U-Net, Detecting Fire Hotspots, Forest Fires, Himawari 8 Satellite, Central Kalimantan.

ABSTRAK

Deteksi dini titik api menjadi sangat penting untuk mencegah penyebaran kebakaran yang lebih luas dan mengurangi dampak lingkungan yang ditimbulkan. Penelitian ini mengusulkan penggunaan model U-Net untuk mendeteksi titik api pada citra satelit Himawari 8, khususnya dengan menggunakan Band 7, yang sensitif terhadap radiasi termal. Data titik api dari MODIS dan VIIRS digunakan sebagai ground truth untuk memvalidasi prediksi model. Model U-Net dilatih menggunakan citra satelit dengan metric binary crossentropy sebagai fungsi loss, dan dievaluasi menggunakan metrik Area Under Curve (AUC), Percentage Correct (PC), Omission Error (OE), dan Commission Error (CE). Hasil pelatihan menunjukkan bahwa model dapat mendeteksi titik api dengan akurasi yang tinggi, dengan AUC mencapai lebih dari 0.85 pada data validasi setelah 300 epoch. Selain itu, PC meningkat dan OE menurun, yang menandakan peningkatan kemampuan model dalam mendeteksi titik api secara akurat. Kesimpulannya, model U-Net yang dikembangkan dalam penelitian ini mampu mendeteksi titik api dengan performa yang baik.

Kata Kunci: U-Net, Deteksi Titik Api, Kebakaran Hutan, Satelit Himawari 8, Kalimantan Tengah.

Implementasi Text Summarization pada Ulasan Aplikasi Mobile JKN Menggunakan TF-IDF dan Cosine Similarity

2025-02-05T07:34:47+00:00

ABSTRACT

Mobile JKN is an application developed by Badan Penyelenggara Jaminan Sosial (BPJS) Kesehatan that was developed for easier access to national health services. As of November 2024, this application has been downloaded 50 million times on the Google Play Store, and about 720 thousand reviews have been given by users. The reviews provided by users who have downloaded the Mobile JKN app are very useful and important for potential users and developers. However, the huge volume of reviews is a challenge in reading them one by one and can cause information overload. Based on the occurring problems, the author will apply text summarization to summarize the reviews of the JKN Mobile application by implementing the Term Frequency-Inverse Document Frequency (TF-IDF) and Cosine Similarity methods. The author added the Maximum Marginal Relevance (MMR) method because the TF-IDF and Cosine Similarity methods cannot produce a summary. Summarization is done by taking the most relevant reviews from among a collection of other reviews. This research resulted with the average Accuracy value of 29.6%, Precision 55.6%, and Recall 39.8%, with the highest value of Accuracy 38.4%, Precision 79.2%, and Recall 42.7%.

Keywords: Cosine Similarity, Text Summarization, TF-IDF, MMR

ABSTRAK

Aplikasi Mobile JKN merupakan aplikasi yang dikembangkan oleh Badan Penyelenggara Jaminan Sosial (BPJS) Kesehatan yang dikembangkan untuk kemudahan akses layanan kesehatan nasional. Per November 2024, aplikasi ini telah diunduh sekitar 50 juta kali di Google Play Store dan sekitar 720 ribu ulasan telah diberikan oleh para pengguna. Ulasan-ulasan yang diberikan oleh pengguna yang telah mengunduh aplikasi Mobile JKN sangat bermanfaat dan penting bagi calon pengguna dan pengembang. Akan tetapi, volume ulasan yang sangat besar menjadi tantangan dalam membacanya satu per satu dan dapat menimbulkan information overload. Berdasarkan permasalahan yang terjadi, maka penulis akan menerapkan text summarization untuk meringkas ulasan-ulasan aplikasi Mobile JKN dengan mengimplementasikan metode Term Frequency-Inverse Document Frequency (TF-IDF) dan Cosine Similarity. Penulis menambahkan metode Maximum Marginal Relevance (MMR) karena metode TF-IDF dan Cosine Similarity tidak dapat menghasilkan ringkasan. Peringkasan dilakukan dengan mengambil ulasan-ulasan yang paling relevan dari antara kumpulan ulasan lainnya. Penelitian ini menghasilkan nilai rata-rata Accuracy 29,6%, Precision 55,6%, dan Recall 39,8% dengan nilai tertinggi Accuracy 38,4%, Precision 79,2%, dan Recall 42,7%.

Kata Kunci: Cosine Similarity, Text Summarization, TF-IDF, MMR

Penggunaan Algoritma LSTM Dengan Mekanisme Attention Untuk Prediksi Harga Cryptocurrency

2025-06-05T05:05:59+00:00

Cryptocurrency, particularly Ethereum, is a digital asset with high price volatility, making price prediction a challenging task for investors. This study aims to develop an Ethereum price prediction model using the Long Short-Term Memory (LSTM) algorithm with an attention mechanism. The dataset includes daily closing prices of Ethereum from November 2017 to October 2024, obtained from Yahoo Finance. The research methodology involves data preprocessing, LSTM model development, and the integration of an attention mechanism to enhance prediction accuracy. Model evaluation was conducted using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE). The results show that the LSTM model with an attention mechanism achieved a lower MAPE of 2.72% and an RMSE of 102.78, demonstrating better adaptability to significant price fluctuations. Meanwhile, the standard LSTM model exhibited greater stability in consistent market conditions, with a MAPE of 3.28% and an RMSE of 115.00. This study contributes to the advancement of cryptocurrency price prediction technology and serves as a reference for future research. Keywords: LSTM, Attention Mechanism, Price Prediction, Ethereum, Cryptocurrency

Perancangan Sistem Informasi Layanan Pengaduan Masyarakat “SORAK” Menggunakan Metode Waterfall

2025-06-05T16:09:38+00:00

This research aims to design an information system called SORAK that facilitates online public complaints and enhances public transparency in complaint management. The research adopts the waterfall method of information system development, consisting of six stages. The main objective of this research is to minimize unresolved complaints through more effective and efficient complaint management. SORAK also provides convenience for the public to submit complaints and track the status of their complaints. Primary data is obtained through interviews with SORAK users, while secondary data is collected through literature review on public complaint management systems. The testing plan for the SORAK application includes functional and performance testing to ensure its proper operation and meet user needs. In conclusion, the author designs SORAK Information System as a public complaint service using the Waterfall method, with a focus on enhancing public transparency. Keywords: Information System, Website, Waterfall, Public Complaints, Public Transparency.