Behavioral and Physiological Neural Network Analyses: A Common Pathway Toward Pattern Recognition and Prediction

Neural Network and Multivariate Analyses: Pattern Recognition in Academic and Social Research

“…That is, when converted to natural logs and classified by the SOM, measures of student performances showed distinctive and highly correlated pattern formations that were entirely absent when the analysis was confined to conventional multivariate statistical procedures. These outcomes are consistent with previous investigations focusing on math performance (Ninness et al, 2005) as well as with findings pertaining to SOM analysis of Congressional members' voting patterns, neurophysiological changes, and diagnostics related to distinguishing benign from malignant cells (Ninness et al, 2012). Ironically, the purpose of employing multivariate analysis and post hoc techniques is to allow researchers to disambiguate findings among and between participants across experimental arrangements.…”

“…Identifying student error patterns allowed us to develop a more efficient remediation strategy during the second experiment within the same paper. A more recent study conducted by the same behavioral software design team applied SOM pattern recognition principles to diversified nonlinear data distributions from behavioral and physiological research, extending previous findings that validate the predictive ability of SOM neural networks (Ninness et al, 2012). Three studies analyzing independent datasets were conducted.…”

“…As mentioned at the beginning of this paper, we (Ninness et al, 2012) previously employed the SOM using data obtained by way of an A-B-A single-subject design to determine the potential coordinated effects of nine variables associated with sympathetic nervous system cardiac activation and various components of the inner ear during a series of cold pressor challenges. Subsequent to SOM analysis, several well-differentiated classifications were identified by the SOM architecture.…”

“…From a direct problem solving perspective, neural networks are congenial with (or complemented by) several statistical algorithms. In the areas of behavioral and social sciences, outcomes conventionally evaluated by factor analysis, principal components analysis, and a wide range of related multivariate procedures may now be evaluated (or at least supplemented) by way of neural network strategies (see Ninness, Lauter, Coffee, Clary, Kelly, Rumph, Rumph, Kyle, & Ninness, 2012, for a discussion). While a comprehensive description of neural networking methodology is well beyond the scope of this paper, we will provide a condensed overview of two commonly employed techniques with a special emphasis on one increasingly popular and exceptionally robust neural network model.…”

“…Unlike supervised neural network systems, the Kohonen SOM "acquires the ability" to classify nonlinear datasets without repetitive inspection of each outcome by the researcher. As described in Ninness et al (2012), rather than successively examining a sequence of training simulations with training data using correction procedures employed by the user, the SOM acts to "self-organize" and systematically minimize the Euclidean distance among input values, producing a constellation of best fit data patterns (see Table 1 regarding Euclidean distance). That is, given that a set of unambiguous patterns may exist within a given chaotic nonlinear dataset, the SOM algorithm isolates and numerically classifies nonlinear patterns that form a visually conspicuous cluster of output values.…”