Prediksi Harga Karbon Menggunakan Pyramidal Recurrent Transformer (PRformer) dan Sentimen Publik Berbasis Intuitionistic Fuzzy Inference System (IFIS)

Mujiono, Edo Priyo Utomo Putro (2025) Prediksi Harga Karbon Menggunakan Pyramidal Recurrent Transformer (PRformer) dan Sentimen Publik Berbasis Intuitionistic Fuzzy Inference System (IFIS). Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6002241001-Mater_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (9MB) | Request a copy

Abstract

Dalam beberapa tahun terakhir, perdagangan karbon telah menjadi instrumen utama dalam upaya untuk mengurangi emisi gas rumah kaca. Namun, adanya fluktuasi harga karbon yang tinggi menjadi tantangan dalam pengambilan keputusan bagi pemangku kepentingan dan investor. Fluktuasi ini dipengaruhi oleh berbagai faktor, termasuk harga komoditas energi, seperti minyak bumi dan batubara, serta berita online. Berita online menjadi sumber informasi tekstual yang penting untuk menggambarkan tren dan peristiwa terkait masalah keuangan, ekonomi, dan politik, sehingga berita online sering digunakan dalam analisis prediksi di berbagai bidang. Prediksi harga karbon bersifat non-linear dan dipengaruhi oleh banyak variabel yang saling terkait, sehingga diperlukan model prediksi yang mampu menangkap fitur temporal multiskala dan hubungan antar-variabel secara efektif. Penelitian ini mengusulkan pendekatan PRformer untuk memprediksi harga karbon dan sentimen publik berbasis Intuitionistic Fuzzy Inference System (IFIS). Berita online diproses melalui Financial BERT (FinBERT) untuk memperoleh probabilitas positif dan negatif karena FinBERT mampu memahami konteks ekonomi dan finansial secara lebih baik daripada analisis sentimen umum. Selanjutnya, untuk mengatasi ketidakpastian dalam sentimen suatu berita, digunakan IFIS yang memetakan probabilitas sentimen menjadi skor representatif dengan mempertimbangkan membership, non-membership, dan hesitation, sehingga meningkatkan keakuratan analisis sentimen. Dengan kombinasi kedua metode ini menjadikan model lebih adaptif terhadap perubahan pasar dan memberikan prediksi harga karbon yang akurat. Pengujian model dilakukan dengan berbagai kombinasi rasio data dan nilai epoch, serta uji hyperparameter. Hasil terbaik diperoleh pada rasio data 75:15:10 dan 50 epoch, dengan nilai MAE sebesar 0.839, MSE 1.183, RMSE 1.087, dan MAPE 1.240%.
=======================================================================================================================================
In recent years, carbon trading has become a key instrument in efforts to reduce greenhouse gas emissions. However, high carbon price fluctuations pose challenges for stakeholders and investors in decision-making. These f luctuations are influenced by various factors, including the prices of energy commodities, such as oil and coal, and online news. Online news is an important source of textual information for describing trends and events related to financial, economic, and political issues, and is therefore frequently used in predictive analytics across various fields. Carbon price predictions are non-linear and influenced by numerous interrelated variables, requiring predictive models that can effectively capture multi-scale temporal features and the relationships between these variables. This study proposes a PRformer approach to predicting carbon prices and public sentiment based on an Intuitionist Fuzzy Inference System (IFIS). Online news is processed through Financial BERT (FinBERT) to obtain positive and negative probabilities because FinBERT is better able to understand the economic and financial context than general sentiment analysis. Furthermore, to address news sentiment uncertainty, IFIS is used, which maps sentiment probabilities into representative scores by considering membership, non-membership, and uncertainty, thereby improving the accuracy of sentiment analysis. The combination of these two methods makes the model more adaptable to market changes and provides accurate carbon price predictions. Model testing was conducted using various combinations of data ratios and epoch values, as well as hyperparameter testing. The best results were obtained with a data ratio of 75:15:10 and 50 epochs, with an MAE of 0.839, MSE of 1.183, RMSE of 1.087, and MAPE of 1.240%.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Harga karbon, Sentimen Publik, PRformer, Finbert, Intuitionistic fuzzy set
Subjects:	Q Science > QA Mathematics > QA39.3 Fuzzy mathematics Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) Q Science > QA Mathematics > QA248_Fuzzy Sets
Divisions:	Faculty of Mathematics and Science > Mathematics > 44101-(S2) Master Thesis
Depositing User:	Edo Priyo Utomo Putro Mujiono
Date Deposited:	01 Aug 2025 06:09
Last Modified:	01 Aug 2025 06:09
URI:	http://repository.its.ac.id/id/eprint/125437

Actions (login required)

View Item