ANN-based predictive analytics of forecasting with sparse data: Applications in data mining contexts

Abstract

Technoeconomics of a business structure exhibit evolving performance attributes as decided by various exogenous and endogenous causative variables. Proposed in this paper is a predictive model to elucidate the forecast performance on such evolving traits in large business structures (like electric power utility companies). The method uses artificial neural network (ANN) based predictive analytics viewed in data mining contexts. Specifically, should the available data be sparse, a method of scarcity removal in the knowledge domain is proposed for subsequent use in the ANN-based data mining exercise. Hence forecast projections on the growth/decay profile across the ex ante regime are determined. Further, for each forecast projection, a cone-of-forecast is suggested toward the corresponding limits (error-bounds) on the accuracy of rules extraction in data mining. Example simulations pertinent to real-world data on the performance of wind-power generation versus wind-speed are presented demonstrating the efficacy of forecasting strategy pursued. Possible shortcomings of the proposals are identified.