Search for collections on Undip Repository

PERBANDINGAN PERFORMA MODEL DEEP LEARNING UNTUK PREDIKSI HARGA CRYPTOCURRENCY

PAMBUDI, Ryo and Nugraheni, Dinar Mutiara Kusumo and Widodo, Aris Puji (2026) PERBANDINGAN PERFORMA MODEL DEEP LEARNING UNTUK PREDIKSI HARGA CRYPTOCURRENCY. Masters thesis, UNIVERSITAS DIPONEGORO.

[thumbnail of cover aja.pdf] Text
cover aja.pdf

Download (179kB)
[thumbnail of Cover_Thesis Ryo Pambudi.pdf] Text
Cover_Thesis Ryo Pambudi.pdf
Restricted to Repository staff only

Download (3MB)
[thumbnail of Bab1_Thesis Ryo Pambudi.pdf] Text
Bab1_Thesis Ryo Pambudi.pdf

Download (2MB)
[thumbnail of Bab2_Thesis Ryo Pambudi.pdf] Text
Bab2_Thesis Ryo Pambudi.pdf

Download (8MB)
[thumbnail of Bab3_Thesis Ryo Pambudi.pdf] Text
Bab3_Thesis Ryo Pambudi.pdf
Restricted to Repository staff only

Download (3MB)
[thumbnail of Bab4_Thesis Ryo Pambudi.pdf] Text
Bab4_Thesis Ryo Pambudi.pdf
Restricted to Repository staff only

Download (11MB)
[thumbnail of Bab5_Thesis Ryo Pambudi.pdf] Text
Bab5_Thesis Ryo Pambudi.pdf
Restricted to Repository staff only

Download (248kB)
[thumbnail of Daftar Pustaka.pdf] Text
Daftar Pustaka.pdf

Download (5MB)

Abstract

Penelitian ini bertujuan untuk membandingkan, mengevaluasi dan menemukan model prediksi harga cryptocurrency yang optimal untuk mengurangi risiko keuangan yang disebabkan oleh volatilitas harga yang tinggi. Penelitian ini membandingkan kemampuan prediksi dari lima model deep learning, yaitu LSTM, GRU, BiLSTM, Transformer, dan Performer, untuk memprediksi harga cryptocurrency dengan akurasi tertinggi di pasar keuangan digital. Tahapan penelitian ini terdiri dari pengumpulan dataset, pra-pemrosesan data, pelatihan model, dan evaluasi model. Dataset yang digunakan dalam penelitian ini, yaitu data harga per menit untuk BTC, ETH, BNB, dan XRP, diperoleh dari Kaggle. Pemrosesan data meliputi normalisasi menggunakan MinMaxScaler dan pembuatan urutan melalui teknik Sliding Window. Untuk memvalidasi setiap model deep learning, empat metrik yang terdiri dari MAE, MSE, RMSE, dan MAPE digunakan untuk evaluasi. Model Transformer menghasilkan hasil terbaik dengan nilai MAPE terendah di semua dataset, terkecil pada BTC sebesar 0,20%, ETH sebesar 0,45%, BNB sebesar 0,29%, dan XRP sebesar 0,36%, menunjukkan akurasi prediksi yang tinggi. BiLSTM menempati peringkat kedua karena secara efektif menangkap ketergantungan temporal dua arah, GRU berada pada peringkat sedang tetapi stabil dalam kinerjanya. sedangkan LSTM dan Performer menghasilkan prediksi yang bervariasi. Hal ini menunjukkan mekanisme self-attention pada Transformer terbukti lebih efektif dalam menangkap dependensi jangka pendek maupun jangka panjang pada data harga cryptocurrency dibandingkan dengan model rekuren konvensional, sehingga menghasilkan prediksi yang lebih akurat. Penelitian ini menawarkan perbandingan komprehensif berbagai model deep learning secara detail, sehingga memungkinkan untuk menemukan model terbaik untuk memprediksi harga cryptocurrency dengan akurasi tinggi. Penelitian inijuga memberikan wawasan berharga untuk pengembangan sistem peramalan harga berbasis deep learning tingkat lanjut di bidang analisis keuangan digital.
Kata kunci: Prediksi Harga Cryptocurrency, LSTM, BiLSTM, GRU, Transformer, Performer

The purpose of this study is to compare, evaluate, and find an optimal price prediction model for cryptocurrency to reduce financial risk caused by high price volatility. This study compares the predictive capabilities of five deep learning models, namely LSTM, GRU, BiLSTM, Transformer, and Performer, to predict cryptocurrency prices with the highest accuracy in the digital financial market. The research process consists of four stages: data collection, pre-processing, model training, and model evaluation. The datasets used in this study, namely minute-by-minute price data for BTC, ETH, BNB, and XRP, were obtained from Kaggle. Data processing included normalization using MinMaxScaler and sequence generation using the Sliding Window technique. To validate each deep learning model, four metrics consisting of MAE, MSE, RMSE, and MAPE were used for evaluation. The Transformer model produced the best results with the lowest MAPE value across all datasets, the smallest for BTC at 0.20%, ETH at 0.45%, BNB at 0.29%, and XRP at 0.36%, indicating high accuracy and generalization. BiLSTM ranked second because it effectively captures bidirectional temporal dependencies; GRU ranked in the middle but demonstrated stable performance, whereas LSTM and Performer produced inconsistent predictions. This indicates that the self-attention mechanism in the Transformer has proven to be more effective at capturing both short-term and long-term dependencies in cryptocurrency price data compared to conventional recurrent models, leading to more accurate predictions. This study offers a comprehensive and detailed comparison of various deep learning models, making it possible to identify the best model for predicting cryptocurrency prices with high accuracy. This study provides valuable insights for the development of advanced deep learning-based price forecasting systems in the field of digital financial analysis.
Keywords: Cryptocurrency Price Prediction, LSTM, BiLSTM, GRU, Transformer, Performer

Item Type: Thesis (Masters)
Uncontrolled Keywords: Prediksi Harga Cryptocurrency, LSTM, BiLSTM, GRU, Transformer, Performer
Subjects: Sciences and Mathemathic
Divisions: Postgraduate Program > Master Program in Information System
Depositing User: ekana listianawati
Date Deposited: 08 Sep 2026 07:51
Last Modified: 08 Sep 2026 07:51
URI: https://eprints2.undip.ac.id/id/eprint/60693

Actions (login required)

View Item View Item