Pattern Modelling in Time-Series Forecasting

Pattern modelling in time-series prediction refers to the process of identifying past relationships and trends in historical data for predicting future values. This paper describes the development of a new pattern matching technique for univariate time-series forecasting. The pattern modelling technique out-performs frequently used statistical methods such as Exponential Smoothing on different error measures and predicting the direction of change in time-series. The paper discusses the prediction results on popular benchmarks and the real US SP index for financial markets. 3 PATTERN MODELLING IN TIME-SERIES FORECASTING INTRODUCTION Univariate time-series prediction is important in several scientific domains. A forecasting procedure may be needed for predicting a single time-dependent variable or predicting several independent variables individually to forecast a dependent variable as in multivariate analysis. Univariate series may be predicted using a range of available tools. De...