A Generic Preprocessing Optimization Methodology when Predicting Time-Series Data

Abstract

A general Methodology referred to as Daphne is introduced which is used to find optimum combinations of methods to preprocess and forecast for time-series datasets. The Daphne Optimization Methodology (DOM) is based on the idea of quantifying the effect of each method on the forecasting performance, and using this information as a distance in a directed graph. Two optimization algorithms, Genetic Algorithms and Ant Colony Optimization, were used for the materialization of the DOM. Results show that the DOM finds a near optimal solution in relatively less time than using the traditional optimization algorithms.