DEVELOPMENT OF DEEP LEARNING-BASED RECOMMENDATION SYSTEM FOR CRYPTOCURRENCY TRADING

• Main Author: Istiqomah, Widad
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/66349
• Institution: Institut Teknologi Bandung

Cryptocurrency are digital currency that are built and secured using a cryptographic-based blockchain. Cryptocurrency has three characteristics, namely decentralization, peer-to-peer, and trustless (Nakamoto, 2008) which are the advantages of cryptocurrencies compared to state currencies. Its advantages and increasingly widespread application bring cryptocurrency to continue to develop positively, both abroad and domestically. Indonesia through BAPPEBTI has determined cryptocurrency as a digital commodity which has been in effect since December 17, 2022. Cryptocurrency as an investment instrument has high-volatility characteristics with high risk-high return on investment. Fundamental knowledge of the cryptocurrency market for optimal profit can be complemented using intelligent systems that have been applied to the states currencies and stock markets. Therefore, a system that can recommend buying and selling decisions on Bitcoin-USD Tether (BTCUSDT) cryptocurrency trading will be developed and researched in this final project. The recommendation system will be built with deep learning approach using Long-Short Term Memory (LSTM) algorithm through the CRISP-DM framework. The model was developed using features of EMA to close ratio and percent change of technical indicators. The performance of the model was evaluated using business (gain and ROI) and non-business (AUC, precision, recall, accuracy) metrics with good results, namely 1.83% gain and 1.7% ROI for the first 7 days of November 2021, 8.55% gain and 0.51% ROI, bigger than buy and hold strategy during November 2021, from the prediction results with 57.7% AUC and 57% accuracy.