]Things to know about the offline development of multiplicative models
Steps/recommendations:
- The difference between additive and multiplicative models is that in the first one, estimation values are formed from the sum of the components, whereas in the latter, we use the PRODUCT() function.
- As a consequence, a multiplicative model is capable to map those occurrences, in which the absence of an input makes the output impossible (e.g. production functions in the agriculture).
- Multiplicative models are needed too, where the single factors are not independent from the others e.g. healthcare models (cf. risk management for hearth diseases)
- We shall add here, that besides the sum- and product-type contractions, estimation functions may have different components: e.g. median, average, standard deviation, quartile, sinus, abs, etc.
- In case of median/average, neural processes can be simulated (cf. activation), where the stimuli spread further only if special unions are formed (cf. yield estimation).
- The precision of the models can be increased arbitrarily through the adequate choice of the estimation function (cf. function-generating). Input-side indicator-generation has a similar effect, which can give any kind of relation between any kinds of attributes. However, the following things need to be kept in mind in such cases: in case of any of two primary indicators, it must be checked whether the ceteris paribus figures can be expected to run (so, what is their meaning).
- On input-side, operations, like raising a given variable to the second power, have no definitive effect on similarity analysis, because this operation does not produce a new ranking-vector.
- Arbitrarily new ranking vectors can be gained by trigonometric functions and even by the ABS() function. Keep in mind to preserve interpretability.
- ...
