For sales and demand forecasting, inventory optimization and inventory planning.

Predictive Analytics

For time series forecasting, regression, curve fitting, classification and clustering.

# Time Series Forecasting Solution for Business

## Raise Forecast Accuracy with Powerful Time Series Forecasting Software

Time series forecasting is a data analysis method that aims to reveal certain patterns from the dataset in an attempt to predict future values. The example of time series data are stock exchange rates, electricity load statistics, monthly (daily, hourly) customer demand data, micro and macroeconomic parameters, genetic patterns and many others.

GMDH Shell is time series forecasting software developed by Geos Research Group on the basis of the classical GMDH algorithm. GMDH (group method of data handling) was developed in the late 1960s by Ukrainian professor Ivakhnenko. The method implies creating of models each one being more complex than the prior one. The coefficients of the regression equation of those models are determined by the least square method. After a model is built, it is applied to separate previously unseen historical data to see if it gives adequate estimates for real historical values. If the approximation error is too high, the model is declined and the next, more complex model is built.

The software is the best I have ever used. What is most impressive, besides the other algorithms, is especially the neural net and time-series forecasting capabilities and the ease with which the formulas can be generated and exported to a spreadsheet for customization.

GMDH Shell inherits GMDH algorithm while making it much more flexible and complex. The tool can apply various approximation methods (linear, polynomial, Gaussian, neural networks) to build a predictive model.

Since the process starts from simple approximations and goes down to more complicated models, the overall time needed to receive a good prediction is generally less than that of many other similar solutions. In addition, GMDH Shell takes the most of your hardware and loads all available cores and CPUs in full, paralleling computations. As the result, building a predictive model from a 200k rows dataset takes about 37 minutes in total – apparently the best result among competitors.

Add here a nice and easy to comprehend interface, a number of predefined samples and a free trial and you’ll get probably the best time series forecasting solution on the market!