Prediction of cryptocurrency prices relatively
accurate remains a formidable challenge due to inherent
volatility associated with it and the absence of traditional
valuation metrics. This research explores the performance of
Long Short-Term Memory (LSTM), Support Vector Machine
(SVM), and a hybrid model of LSTM+SVM for this complex
task. LSTM has demonstrated potential in capturing short-term
price fluctuations, while the hybrid model aims to combine the
strengths of temporal dependencies of LSTM and pattern
recognition of SVM. To evaluate the models’ performance,
comprehensive evaluation framework has been employed,
considering generalization ability of the models, robustness,
computational efficiency, and interpretability. Historical daily
price data for five leading cryptocurrencies, Ethereum, Solana,
BNB, Tether, and Bitcoin was collected from 2020 to 2024.
This data was used to evaluate the model performance of
LSTM, SVM, and a hybrid model of them, using metrics such
as R-Square, Root Mean Square Error (RMSE), and Mean
Absolute Error (MAE). The findings from the study indicate
that LSTM generally outperformed both SVM and the hybrid
model in terms of these evaluation metrics. Moreover, the
hybrid model demonstrated competitive performance,
particularly when considering its statistical significance and
ability to generalize across different volatile conditions. While
SVM model has potential, it requires meticulous
hyperparameter tuning and feature engineering to reach optimal
performance. This research offers a comparative analysis of
machine learning models for cryptocurrency price forecasting,
detailing the strengths and limitations of LSTM, SVM, and
hybrid approaches. The insights provided are valuable for both
practitioners and researchers. Future studies could explore more
advanced hybrid architectures considering different algorithms,
incorporate additional data sources, and assess how varying
market conditions may affect model performance.