Feature Extraction using Singular Spectrum Analysis: Characterizing Dominant Modes for Time Series Forecasting

Abstract

This study explores the application of Singular Spectrum Analysis (SSA) for feature extraction from the dominant modes of an industry sector. These modes are hypothesized to encapsulate the underlying market trends and cycles, offering an enhanced understanding of stock price dynamics. The methodology involves identification of dominant modes of historical stock price data from leading semiconductor companies, and applying Singular Spectrum Analysis (SSA) to identify and isolate the relevant features contributing to price dynamics. Finally, the features extracted are used to forecast a new time series in the same sector using Elastic Net Regression. The forecasting evaluation metrics indicates lower error rates and high predictive accuracy.