Demantra spectrum product has a set of various mathematical models geared up for capturing various demand patterns. It utilizes Bayesian modeling technique to combine the result of forecast generated by individual 15 mathematical models. The final forecast doesn't merely represents results based on prediction done by just selecting a mathematical model which can best fit the historical pattern of demand. Demantra's patented Bayesian modeling forecasting engine, instead captures the qualitative prediction done by multiple models and thus resulting in one of the best forecasting results in industry.
Oracle's Demantra Demand Management module only provides 9 basic mathematical models and 6 configurable Causal Factors, rest of the 6 advanced statistical models and flexibility of creating unlimited Casual Factors are given away with Advanced Forecasting & Demand Modeling (AFDM) module.
Demand Management module's forecast library has following set of models for usage:
1. Regression
2. Transformation Model(log)
3. Regression for Intermittent
4. Holt
5. Croston for Intermittent
6. Combined Transformation Model(elog)
7. Multiplicative Monte Carlo Regression(CMReg)
8. Integrated Causal Exponential Model(BWint)
9. Auto & Linear Regression
Let's try to get a feel of what are these models and how they work ?
1. Regression: These models are statistical models which are capable enough of describing the variation(trend/pattern) one or more variable(s), based on the variation of one or more other variable(s). Inferences based on this kind of methodology are known as Regression analysis.
e.g. Variation of Demand of a product in market based on time variation.
2. Transformation Model(log): Log transformation model utilizes the log function which squeezes the large values of data together and stretches the small values apart, thus leading to correction data issues like skewed data, outliers and unequal variation.
e.g. Normally demand data has various such problems, the log transformation model tries to minimize those.
3. Regression for Intermittent: This model uses regression analysis for intermittent kind of data. There are parameters related to defining intermittent part of data in the application.
4. Holt: An extension of exponential smoothing can be used when time-series data exhibits a linear trend.
5. Croston for Intermittent: Croston’s Intermittent Model is specifically designed to deal with sporadic demand (no seasonality) with a two-step process. Croston’s Intermittent model recognizes both : the demand size and the demand occurrence.
.. to be continued
