Forecasting Tools in Practical Applications: Selection and Evaluation Methodology

Abstract

We call a set of programs a Prediction Tool (PT) that can be used to solve a particular applied prediction problem, for example, predicting the volumes of traffic under consideration at certain points in the future. The goal may also be a forecast for the network administrator. We analyze the information on the input data used for prediction and the choice of the predictors to be used among a set of predictors.The paper analyzes procedures for choosing a predictor during implementation of prediction online scheme.The predictability properties of random sequences, and the required and achievable accuracy based on estimating the conditional probability of prediction over past history results. Although some of these issues have been considered in sufficient detail in the literature, for example, the analysis of predictability measures, accuracy metrics, however, as will be shown, they are more focused on the problems of constructing specific prediction algorithms rather than focus on the choice of existing predictor from a given predictor set.It is shown how the specified properties of sequences and probability estimates affect the quality of the choice of predictors. Based on this analysis, a rule for choosing a predictor based on the results of previous (potential) predictions is formulated.