Genetic Algorithm for Feature Selection Applied to Financial Time Series Monotonicity Prediction: Experimental Cases...

Area of Science:

• Quantitative Finance
• Computational Finance
• Machine Learning Applications

Background:

• Investors historically predict financial asset values; cryptocurrencies are now significant assets.
• Machine learning (ML) is increasingly used for investment automation and decision support.
• Accurate prediction of financial time series movement direction is crucial for investment strategies.

Purpose of the Study:

• To predict the daily movement direction of financial time series using ML classification models.
• To apply data preprocessing and feature selection techniques for enhanced prediction accuracy.
• To evaluate model performance on Bitcoin, Ibovespa, and Vale time series data.

Main Methods:

• Collection of financial time series data for Bitcoin, Ibovespa, and Vale.
• Application of data preprocessing techniques tailored for financial time series.
• Utilizing genetic algorithms for feature selection to identify optimal input attributes.
• Training and testing various machine learning classification models.

Main Results:

• The best prediction models achieved Area Under the ROC Curve (AUC) scores of 0.61 for Bitcoin, 0.62 for Ibovespa, and 0.58 for Vale.
• Feature selection using genetic algorithms generally improved model performance across the tested assets.
• Financial time series represented as percentage variations showed good predictive performance with fewer attributes.

Conclusions:

• Feature selection is a valuable technique for enhancing the performance of ML models in financial time series prediction.
• Percentage variation of time series data can yield effective predictions, even with a reduced set of features.
• The study demonstrates the potential of ML and genetic algorithms for predicting the direction of financial asset movements.