Prediksi Harga Bitcoin dengan Support Vector Regression dan Long Short-Term Memory Berbasis Indikator Makroekonomi dan Google Trends

Irsyadi, Wildan Nazhif (2026) Prediksi Harga Bitcoin dengan Support Vector Regression dan Long Short-Term Memory Berbasis Indikator Makroekonomi dan Google Trends. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5003221109-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Bitcoin adalah aset kripto dengan volatilitas harga yang sangat tinggi, dipengaruhi oleh sentimen pasar, kebijakan moneter, dan aktivitas pencarian global, sehingga prediksi harganya menjadi tantangan sekaligus kebutuhan nyata bagi investor. Penelitian ini membandingkan tiga metode peramalan, yaitu Autoregressive Integrated Moving Average (ARIMA), Support Vector Regression (SVR), dan Long Short-Term Memory (LSTM), dengan dua pendekatan pemodelan: non-exogenous yang hanya menggunakan data historis harga, dan exogenous yang menambahkan variabel volume perdagangan, indikator makroekonomi (Federal Funds Rate dan Consumer Price Index (CPI) Amerika Serikat), serta indeks Google Trends dengan kata kunci Bitcoin, Bitcoin Mining, Blockchain, dan Ethereum dalam lingkup global (Worldwide). Data bulanan Januari 2016 hingga Desember 2025 digunakan dengan pembagian in-sample Januari 2016 hingga Desember 2024 dan out-sample Januari hingga Desember 2025. Pemilihan lag variabel eksogen dilakukan melalui Cross Correlation Function (CCF), yang menghasilkan lag 0 untuk volume perdagangan dan seluruh variabel Google Trends, lag 1 untuk suku bunga AS, serta lag 6 untuk CPI. SVR menggunakan kernel Radial Basis Function (RBF) dengan strategi MultiOutputRegressor, sedangkan LSTM menggunakan optimizer AdamW dengan fungsi loss Huber dan strategi recursive one-step log-return. Evaluasi kinerja model menggunakan Root Mean Square Error (RMSE) dan Mean Absolute Percentage Error (MAPE). Hasil penelitian menunjukkan bahwa LSTM Eksogen adalah model terbaik dengan MAPE 4,73% dan RMSE 6.085,86 pada data out-sample, mengungguli ARIMA(0,1,1) dengan MAPE 9,54%, SVR tanpa eksogen dengan MAPE 15,15%, LSTM tanpa eksogen dengan MAPE 6,78%, dan SVR Eksogen dengan MAPE 15,03%. Berdasarkan model terbaik ini, peramalan harga Bitcoin untuk Januari hingga Juni 2026 dilakukan dengan dua skenario: statis (nilai eksogen konstan) dan dinamis (proyeksi eksogen menggunakan Auto-ARIMA). Kedua skenario memproyeksikan tren penurunan harga. Pada skenario statis, harga Bitcoin diproyeksikan turun dari 87.462,02 USD pada Januari 2026 menjadi 72.326,41 USD pada Juni 2026. Sementara itu, skenario dinamis menghasilkan proyeksi yang lebih rendah dan adaptif terhadap kondisi makroekonomi serta sentimen pasar, yaitu menurun dari 87.462,02 USD pada Januari 2026 menjadi 67.180,28 USD pada Juni 2026.
=======================================================================================================================================
Bitcoin is a cryptocurrency asset with extremely high price volatility, influenced by market sentiment, monetary policy, and global search activity, making accurate price prediction both a genuine challenge and a practical necessity for investors. This study compares three forecasting methods, namely Autoregressive Integrated Moving Average (ARIMA), Support Vector Regression (SVR), and Long Short-Term Memory (LSTM), across two modeling approaches: a non-exogenous approach using only historical price data, and an exogenous approach incorporating trading volume, macroeconomic indicators (Federal Funds Rate and US Consumer Price Index/CPI), and Google Trends indices with keywords Bitcoin, Bitcoin Mining, Blockchain, and Ethereum, all collected at a global (Worldwide) scope. Monthly data from January 2016 to December 2025 were used, split into an in-sample period of January 2016 to December 2024 and an out-sample period of January to December 2025. Exogenous variable lags were selected via Cross Correlation Function (CCF), resulting in lag 0 for trading volume and all Google Trends variables, lag 1 for the US Federal Funds Rate, and lag 6 for CPI. SVR used the Radial Basis Function (RBF) kernel with a MultiOutputRegressor strategy, while LSTM utilized the AdamW optimizer with Huber loss and a recursive one-step log-return strategy. Model performance was evaluated using Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE). Results show that the Exogenous LSTM model achieved the best performance with a MAPE of 4.73% and RMSE of 6,085.86 on out-sample data, outperforming ARIMA(0,1,1) at 9.54% MAPE, SVR at 15.15%, LSTM at 6.78%, and Exogenous SVR at 15.03%. Based on this best model, Bitcoin price forecasting for January to June 2026 was conducted under two scenarios: static (exogenous variables held constant) and dynamic (exogenous variables projected using Auto-ARIMA). Both scenarios project a downward price trend. In the static scenario, the price is projected to decline from USD 87,462.02 in January 2026 to USD 72,326.41 in June 2026. Meanwhile, the dynamic scenario yields a lower and more adaptive projection in response to macroeconomic conditions and market sentiment, decreasing from USD 87,462.02 in January 2026 to USD 67,180.28 in June 2026.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Bitcoin, Support Vector Regression (SVR), Long Short-Term Memory (LSTM), Peramalan Harga, Makroekonomi, Google Trends, Price Forecasting, Macroeconomics
Subjects:	H Social Sciences > HA Statistics > HA30.3 Time-series analysis Q Science > QA Mathematics > QA276 Mathematical statistics. Time-series analysis. Failure time data analysis. Survival analysis (Biometry) T Technology > T Technology (General) > T174 Technological forecasting T Technology > T Technology (General) > T57.8 Nonlinear programming. Support vector machine. Wavelets. Hidden Markov models.
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	Wildan Nazhif Irsyadi
Date Deposited:	03 Aug 2026 02:30
Last Modified:	03 Aug 2026 02:30
URI:	http://repository.its.ac.id/id/eprint/139869

Actions (login required)

View Item