Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Financial Time-Series Forecasting: Towards Synergizing Performance And Interpretability Within a Hybrid Machine Learning Approach

Shun Liu, Kexin Wu, Chufeng Jiang, Bin Huang, Danqing Ma

TL;DR

This study investigates Bitcoin price forecasting by integrating time-series preprocessing with both simple and complex models to balance interpretability and accuracy. It compares Linear Regression (OLS and Lasso), Decision Tree, and LSTM variants, using Holt-Winters as a baseline and emphasizing decomposition and autocorrelation analyses. The key finding is that Lasso regression delivers the best predictive performance while maintaining interpretability, whereas LSTM struggles to converge on this dataset. The work offers a practical framework for interpretable forecasting in volatile crypto markets and suggests extending the approach to other assets and real-time data scenarios.

Abstract

In the realm of cryptocurrency, the prediction of Bitcoin prices has garnered substantial attention due to its potential impact on financial markets and investment strategies. This paper propose a comparative study on hybrid machine learning algorithms and leverage on enhancing model interpretability. Specifically, linear regression(OLS, LASSO), long-short term memory(LSTM), decision tree regressors are introduced. Through the grounded experiments, we observe linear regressor achieves the best performance among candidate models. For the interpretability, we carry out a systematic overview on the preprocessing techniques of time-series statistics, including decomposition, auto-correlational function, exponential triple forecasting, which aim to excavate latent relations and complex patterns appeared in the financial time-series forecasting. We believe this work may derive more attention and inspire more researches in the realm of time-series analysis and its realistic applications.

Financial Time-Series Forecasting: Towards Synergizing Performance And Interpretability Within a Hybrid Machine Learning Approach

TL;DR

This study investigates Bitcoin price forecasting by integrating time-series preprocessing with both simple and complex models to balance interpretability and accuracy. It compares Linear Regression (OLS and Lasso), Decision Tree, and LSTM variants, using Holt-Winters as a baseline and emphasizing decomposition and autocorrelation analyses. The key finding is that Lasso regression delivers the best predictive performance while maintaining interpretability, whereas LSTM struggles to converge on this dataset. The work offers a practical framework for interpretable forecasting in volatile crypto markets and suggests extending the approach to other assets and real-time data scenarios.

Abstract

In the realm of cryptocurrency, the prediction of Bitcoin prices has garnered substantial attention due to its potential impact on financial markets and investment strategies. This paper propose a comparative study on hybrid machine learning algorithms and leverage on enhancing model interpretability. Specifically, linear regression(OLS, LASSO), long-short term memory(LSTM), decision tree regressors are introduced. Through the grounded experiments, we observe linear regressor achieves the best performance among candidate models. For the interpretability, we carry out a systematic overview on the preprocessing techniques of time-series statistics, including decomposition, auto-correlational function, exponential triple forecasting, which aim to excavate latent relations and complex patterns appeared in the financial time-series forecasting. We believe this work may derive more attention and inspire more researches in the realm of time-series analysis and its realistic applications.
Paper Structure (13 sections, 4 equations, 7 figures, 1 table)

This paper contains 13 sections, 4 equations, 7 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methodology
  3. EDA and Data Preprocessing
  4. Decomposition
  5. Autocorrelation Function(ACF)
  6. Time-Series Forecasting
  7. Experiment
  8. Preprocessing
  9. Protocol and Metrics
  10. Candidate Models
  11. Results
  12. Discussion
  13. Conclusion

Figures (7)

  • Figure 1: Overview of the closing prices in the collected dataset.
  • Figure 2: Decomposition of historical prices, which is consisted of four components: 'trend' comopnent deliver a long-term picture of the prices; 'seasonality' identifies the regular pattern for a fixed amount of time range; 'random' component address the existence of noises or fluctuations in the raw dataset, hence introduce some uncertainty to the model, this stage conjecture samples are independent and identically distributed(i.i.d); 'observed' component reflects real marketing data, which integrates the rest three components to formulate a comprehensive depictions.
  • Figure 3: Autocorrelation function(ACF) and partial autocorrelation function(PACF) between observed time series and lagged time series.
  • Figure 4: Rolling factors(mean, std.) of bitcoin closing prices in the normal scale(left) and log scale(right).
  • Figure 5: Multi-layer LSTM architecture. Unlike conventional LSTM model, we concatenate more LSTM layers in the single model, and introduce dropout regularization to avoid overfitting, which is essential to the financial time-series forecasting.
  • ...and 2 more figures