Short-Term Price Forecasting of Emerging Tokens: A Time Series Categorization and TFT Approach

Articles in Press

Document Type : Original Article

Authors

1 Software Department, Faculty of Computer Engineering, Yazd University, Yazd, Iran

2 Department of Computer Science New York Institute of Technology Vancouver, BC, Canada

Abstract

A novel approach for enhancing short-term price prediction accuracy of emerging cryptocurrency tokens is presented. By addressing the challenges of limited historical data and high volatility through time series categorization, this method categorizes financial time series into distinct subseries based on shared behavioral patterns. For each category, Temporal Fusion Transformers (TFTs) are used to forecast the next step. A data augmentation technique is proposed to combat limited data, particularly when increasing the number of categories. This technique leverages time series data from multiple cryptocurrencies to enrich the training data, ensuring robust model training and improved predictive power. The methodology is tested on two emerging tokens, Notcoin and Dogs. Results demonstrate that the integration of time series categorization, TFT models, and data augmentation significantly improves short-term price forecasting accuracy. In simulated spot trading, the proposed method achieved a 2.45% higher return compared to the baseline TFT approach and a significant improvement compared to the baseline LSTM approach based on initial investment, without using leverage or futures contracts. These findings have valuable implications for traders and investors seeking to make informed decisions in the emerging token market.

Keywords

TABRIZ JOURNAL OF ELECTRICAL ENGINEERING

Articles in Press, Corrected Proof
Available Online from 24 August 2025

Files

Share

How to cite

Statistics