Neural networks to classify employees for tax purposes

Abstract

Neural network models are compared with logistic regression models to assess their ability to predict federal court judgments in cases classifying workers as employees or independent contractors for tax purposes. Such classification is highly dependent upon the subjective evaluation of certain determining factors. The neural network approach was found to provide a viable alternative for making this prediction. A second experiment compared the predictions of neural network and logistic regression models with those of human novices and experts. It was found that the neural network and logistic regression predictions were superior to those of both human novices and experts. Finally, simple linear regression models were compared with artificial neural network models as well as with human evaluators. The findings were similar to those of the first two experiments.