Online sequential pattern mining and association discovery by advanced artificial intelligence and machine learning techniques

Shian-Chang Huang,Chei-Chang Chiou,Jui-Te Chiang,Cheng-Feng Wu

DOI: https://doi.org/10.1007/s00500-019-04100-5

IF: 3.732

2019-05-30

Soft Computing

Abstract:With the advances in information science, vast amounts of financial time series data can been collected and analyzed. In modern time series analysis, sequential pattern mining (SPM) and association discovery (AD) are the most important techniques to predict the future trends. This study aims at developing advanced SPM and AD for financial data by cutting edge techniques from artificial intelligence and machine learning. The nonlinearity and non-stationarity of financial time series dynamics pose a major challenge for SPM and AD. This study employs time–frequency analysis to extract features for SPM. Then, a sparse multi-manifold clustering (SMMC) is used to partition the feature space into several disjointed regions for better AD. Finally, local relevance vector machines (RVMs) are employed for AD and perform the forecasting. Different from traditional methods, the novel forecasting system operates on multiple resolutions and multiple dynamic regimes. SMMC finds both the neighbors and the weights automatically by a sparse solution, which approximately spans a low-dimensional affine subspace at that point. RVM, the Bayesian kernel machines, can produce parsimonious models with excellent generalization properties. Taking multiple time series data from financial markets as an example, the empirical results demonstrate that the proposed model outperforms traditional models and significantly reduces the forecasting errors. The framework is effective and suitable for other time series forecasting.

computer science, artificial intelligence, interdisciplinary applications