Estimation of Appropriate Forecasts using Neural Network Learning of Previous Sales Data

Abstract

A forecast is an estimate of the level of demand to be expected for the particular product or several products for some period of time in the future. It can be said that forecast is an educated guess; certainly is should contain as little error as is humanly possible. To make a forecast more meaningful, it should be in turns of units to be planned or scheduled, and it should cover a time period at least as long as the period of time required to make a decision and to put that decision into effect .The nature of product and its demand pattern affect the type of forecast to be made and the time period which must be covered or spanned; like demand may be constant, may have cyclic variations, long term upward trend etc.. Generally we find that forecast can be done by subjective opinion, market research, using certain index or can be computed using averages or some combination methods. Most of the forecaster assumes that use of statistical methods applied to the past data is a realistic way of forecasting future demands. New researchers use the bass model and other life cycle assessment models of the product to forecast the demand but these require a similar product life cycle to compute certain constants which again weaken the forecast. We think that the possibility of getting the almost exact forecast depends on the process (may be statistical or any) that are used to forecast, and these processes are selected by the forecaster; this means that these all forecasts basically dependant on the experiences of the forecaster and the knowledge regarding the trend of the market, and also the ability of the forecaster to predict the remaining life cycle of the product. This is the basic region that we get the values which, many times, are far from actual one. Considering the requirements of forecasts i.e. a huge knowledge about the product whose demand is to be forecasted and certainty of irrelevance or missing data, the author experiments on a new tradition of forecasting and uses an artificial neural network technology to forecast the future demand and represent some of them in this paper. This technology uses mathematical formulations to model nervous system operations and is used to learn pattern and relationships in data. There is verity of software available to model the neural network but among those the author fined more convenient software which is a add-ins of MS EXEL known as Neuro XL Predictor which is very simple to use as it is a part of Microsoft Excel spreadsheets.