An Improved Variable Kernel Density Estimator Based on L2 Regularization

The nature of the kernel density estimator (KDE) is to find underlying probability function (p.d.f) for a given dataset. key training KDE determine optimal bandwidth or Parzen window. All data points share fixed (scalar univariate and vector multivariate KDE) in (FKDE). In this paper, we propose an improved variable (IVKDE) which determines each point dataset based on integrated squared error (ISE) criterion with L2 regularization term. An effective optimization algorithm developed solve objective function. We compare estimation performance IVKDE FKDE VKDE ISE without four distributions. experimental results show that obtains lower errors thus demonstrate effectiveness IVKDE.