Forecasting is the process of making predictions of the future based on past and present, historical data and most commonly by analysis of trends. It is a systematic attempt to understand the future performance of a business or the behavior of a system. The understanding of the influencing variables, is an important component to improve the quality of the forecast.
Is not possible to remove the error from future decisions, but a correct approach to the forecast, reduces the risk allowing better informed decisions and improving the resilience respect unforeseen events. The forecasts must be coherent with the nature of the decision or control logic to support and the available data.
Concerning the models, it is possible to distinguish among qualitative and quantitative methods. The former are subjective, based on the opinion of experts; they are used when past data is not available. The latter create a relation, described by a function, between past and future data. They are usable when past data information is available and it can be assumed that some old patterns still are expected in the future.
ACT OR's products are for:
Revenue management and price optimization is a core-activity in any market oriented company operating in different industries (Retails, Fast Fashion, Chain Store, Car Rentals, Web Business, etc.). The goal of our suite is to provide ...
If the inventory management, balancing stock investments and service level are important in your business, ACT OR is proud to provide you an advanced predictive and optimization set of products to reach superior results ...
The products in this suite enablean effective management of logistics, transportation and supply chain by dynamic simulation and math-optimization technologies.
Warehouses and distribution centers benefit by ...
Between quantitative methods, it is possible to further split them between statistical and non-statistical methods. Statistical methods include for example the Moving Average, the Exponential Moving Average, simple Exponential Smoothing, ARMA and ARIMA, Holt Winters, etc. All these methods rely on some assumptions on the data that should be satisfied in order to obtain satisfactory results with their application. These methods typically decompose the series between the trend, the seasonality and the remainder.
On the other hand, non-statistical methods include all the Machine Learning framework within the Artificial Intelligence field. In this field many algorithms can be found, like Neural Networks, Support Vector Machines, Radial Basis Functions, Trees and Forests, etc. All the algorithms belonging to this field refer to Data Mining. They do not explicitly need that data satisfies some assumptions; actually, these algorithms try to mine data in order to find patterns and behaviors to be used for forecasting.
All the above methods belong to supervised learning, which describe the technique used by the tool during its training phase. These methods try to emulate the human brain, which learn first deducing a relation between effect and cause and the can extend this law in order to generalize the phenomenon. For this reason, data provided to this tool must contain both the cause, the input, and the effect, the output.
There are other methods that do not need the input-output pair in order to be trained. These methods, like all the clustering techniques, are catalogued in the non-supervised learning. These methods try to find similarities among data collecting them in groups, accordingly to the features selected to describe them.
Before the application of any of the forecasting methods, the analyst should perform an analysis of the time series in order to verify if the assumptions underlying each method are met in order to correctly use the most appropriate approach.
Finally, clustering techniques have been shown to be a powerful tool in order to identify a customer segmentation looking at their behavior through historical data. This may lead the marketing/sales functions of a company to differentiate and address specialized promo to the different customer segments.
APPLICATION EXAMPLES & BENEFITS
There is a wide usage of forecasting methods. They can be used, for example:
to predict the sales of a product in order to optimize the inventory in the warehouse or plan a production
to predict the booking on a certain web-site
to predict the needs of energy or any other commodity
to predict the needs of product at store level to manage the replenishment
to predict failures for maintenance purposes
to understand the effect of promotions ...