Turning Point Induced Knowledge Forecasting under Uncertainties (TrIK)

Abstract

Time series forecasting is used in many applications like price prediction, stock trend analysis, etc., This forecasting is dependent on multiple variables, which have an impact on contextual uncertainties. The inability to handle contextual uncertainty, induces incorrect predictions, leading to crucial business decision-making failures. This further restricts the abilities of forecasting and decision-making in a dynamic and partially observable real-life environment. This paper proposes a Turning Points-based approach along with shoulder impact processing. This imparts uncertainties in learning and handles such scenarios more scientifically and gracefully. Turning points are predicted based on probable impending surprise. These turning points are forecasted based on uncertainties found in the variables contributing to sudden out-of-the-pattern changes in the next cycle of forecasting. The shoulder impact areas are classified based on trend changes to improve decisions. Multivariate analysis, fuzzy time series analysis, and contextual impact determination assist to identify such uncertainties and inter-dependencies for improved forecasting. These points support trend forecasting instead of forecasting at an instance. This helps in many applications including decision-making regarding pricing, storing, and distribution of short-span, medium-span, and long-term span perishable commodities. TrIK can be extended to multiple time series forecasting applications to reduce wastage and economic loss.