Predicting Organizational Performance Using Hybrid Models Of Pca And Neural Networks

Abstract

In this study proposed a new combined Principal Components Analysis/ Neural Networks model for predicting organisation performance, Principal Components Analysis assists in solving the dimensionality limitation by transforming data in such a manner as to provide raw data sets, and Neural Networks assists in defining the non-linear relationships amongst the various predictor variables by defining non-linear relationships between each predictor variable, thus resulting in improved accuracy of the resulting prediction model. The results from this study also indicate that the new hybrid predictive model developed in this study has produced more precise predictions than previously published empirical predictions due to the increased accuracy of predictions from traditional methods (R^2 = 0.938; 93.8%). Additionally, the combination of Principal Components Analysis and Neural Networks provides a comprehensive methodology for reducing overfitting and complexity, thereby increasing RMSE (0.154) and MAE (0.116). Furthermore, by combining Neural Networks with Principal Components Analysis, were able to decrease the number of input features from 20 to 8; this results in (a) increased computational efficiency (32%); and (b) faster convergence. These results demonstrate that the combination of Principal Components Analysis and Neural Networks provides an alternative means of accurately predicting organisational performance, without increasing computational times, compared to traditional methods. Therefore, organisations may wish to use the models developed in this study to assist with forecasting their respective future performance, in order to enhance operational policy development.