A new approach to applying feedforward neural networks to the prediction of musculoskeletal disorder risk

A new and improved method to feedforward neural network (FNN) development f or application to data classification problems, such as the prediction of l evels of low-back disorder (LBD) risk associated with industrial jobs, is p resented. Background on FNN development for data classification is provided along with discussions of previous research and neighborhood (local) solut ion search methods for hard combinatorial problems. An analytical study is presented which compared prediction accuracy of a FNN based on an error-bac k propagation (EBP) algorithm with the accuracy of a FNN developed by consi dering results of local solution search (simulated annealing) for classifyi ng industrial jobs as posing low or high risk for LBDs. The comparison demo nstrated superior performance of the FNN generated using the new method. Th e architecture of this FNN included fewer input (predictor) variables and h idden neurons than the FNN developed based on the EBP algorithm. Independen t variable selection methods and the phenomenon of 'overfitting' in FNN (an d statistical model) generation for data classification are discussed. The results are supportive of the use of the new approach to FNN development fo r applications to musculoskeletal disorders and risk forecasting in other d omains. (C) 2000 Published by Elsevier Science Ltd. All rights reserved.