Prediction of mechanical lower back pain for healthcare workers using ann and logistic regression models

Abstract

The aim of this study is the comparison of predictive capabilities of logistic regression (LR) model and artificial neural network (ANN) to predict chronic mechanical lower back pain (MLBP) for healthcare workers in North Cyprus. For this purpose, the dataset has been obtained from Near East University (NEU) Hospital healthcare employees after obtaining approval from the ethics committee. Since this work was defined as exploratory, stepwise regression methods were considered to be the most appropriate and therefore, the Forward Selection and Backward Elimination methods were compared to find the proper binary LR model by using Likelihood ratio test. In order to obtain accurate results, two ANN models (ANN_1 and ANN_2) were used in this study. The main different of these two models was the number of processing elements. In the both models the Levenberg–Marquardt (LM) algorithm, as one of the most common and fastest back-propagation training algorithms was used in this study. The predictive capabilities of the binary logistic regression and ANNs was evaluated by specificity, sensitivity, accuracy rates and area under the ROC curve. The comparison results show that ANN performs better than the logistic regression model for prediction of chronic MLBP for healthcare workers. However, two models are biologically acceptable, too.