Pattern Layer Reduction for a Generalized Regression Neural Network by Using a Self–Organizing Map

Abstract

In a general regression neural network (GRNN), the number of neurons in the pattern layer is proportional to the number of training samples in the dataset. The use of a GRNN in applications that have relatively large datasets becomes troublesome due to the architecture and speed required. The great number of neurons in the pattern layer requires a substantial increase in memory usage and causes a substantial decrease in calculation speed. Therefore, there is a strong need for pattern layer size reduction. In this study, a self-organizing map (SOM) structure is introduced as a pre-processor for the GRNN. First, an SOM is generated for the training dataset. Second, each training record is labelled with the most similar map unit. Lastly, when a new test record is applied to the network, the most similar map units are detected, and the training data that have the same labels as the detected units are fed into the network instead of the entire training dataset. This scheme enables a considerable reduction in the pattern layer size. The proposed hybrid model was evaluated by using fifteen benchmark test functions and eight different UCI datasets. According to the simulation results, the proposed model significantly simplifies the GRNN’s structure without any performance loss.

A Novel Method Based on Neural Networks for Designing Internal Coverings in Buildings: Energy Saving and Thermal Comfort

“…The neuronal model of generalized regression employed in this research work (GRNN) was proposed and developed by Specht in 1991 [27,28]. It possesses the desirable property of not requiring any iterative training, that is, it can approximate any arbitrary function between input vectors (inputs) and output vectors (outputs), taking the estimation of the function directly from the training data.…”

“…In this sense, the GRNN model relies on nonlinear regression theory. It is, in essence, a method to estimate a function f (x, y) only through the training set, so that the joint probability function, which is unknown, is estimated using the Parzen estimator [27,28]. To do this, we must first define the following distances between "x" and "y".…”

Margin-based approach for outlier detection of industrial design data using a modified general regression neural network

“…In other published work, supervised, semi-supervised, and unsupervised methods have all been used for outlier detection (Hodge and Austin, 2004; Singh and Cantt, 2012; Xu et al , 2018). Among the unsupervised methods, two commonly used approaches to detect outliers are: statistical models to estimate a probability density function (p.d.f) using methods such as Parzen window (Parzen, 1962; Mussa et al , 2015; Wang et al , 2019); and ANNs that establish the relationship between variables as a regression model such as the GRNN (Specht, 1991; Kartal et al , 2018; Wang et al , 2019). Unsupervised methods can learn directly from industrial design datasets, offering clear economic benefits, and use these models to validate data by detecting outliers.…”