Inductive Solutions, Inc.
380 Rector Place, Suite 4A, New York, New York 10280

Email  Telephone: +1 (212)945.0630   



Products and Services
Software Products
Recommended Books

Bibliography and White Papers
Free Downloads






KernelNet is a statistical pattern recognition and forecasting tool based on non-parametric multivariate kernel density regression, a method that is based on estimating the empirical probability density of the underlying data.  (The "Kernel" is actually a specific probability density function.)  In the neural network community, this method is also known as a General Regression Neural Network.  TKernelNet is based on a very fast C/C++ code and is limited by dynamic memory. It requires 64-bit implentations of Windows.

KernelNet Features

  • More efficient than other neural network techniques and usually more accurate
  • Uses principal component analysis to automatically reduce the input space of high dimensional problems (using a variation of a technique called Sliced Inverse Regression)
  • Provides the Gausian and Epicheknikov kernels


  • Easy-to-Learn and Easy-to-Use Excel Spreadsheet User Interface
  • Computation is very fast
  • Computes standard statistical metrics for goodness-of-fit and forecasting (including R-SQUARED, ADJUSTED R-SQUARED, STANDARD ERROR, DURBIN-WATSON, RMS ERROR, THEIL INEQUALITY COEFFICIENT)
  • The KernelNet C/C++ Library is available for further customization


The standard single user license is for Microsoft Windows.  Other licensing plans for other platforms are also available. Contact us about versions for other operating systems (such as Linux or Solaris), about site licenses, or about academic discounts.

The kernel density regression method is much more efficient than other neural network techniques and usually more accurate.  For example, the following chart compares the kernel density regression and the classical back propagation neural network training algorithm over a set of data inputs for a simple forecasting problem:

The back propagation algorithm executed through 100,000 epochs (complete cycles through the entire data set); the kernel density regression algorithm required only 1 epoch and yields higher accuracy.  Like other training algorithms, accuracy is influenced by certain parameters.  In kernel density regression, accuracy is controlled through the bandwidth parameter. 


(c) 2015 Inductive Solutions, Inc. All rights reserved.