An adaptive neuro-fuzzy inference system for predicting the risks of low back disorders due to manual material lifting jobs

• The study builds and tests several ANFIS models on a seminal Marras data set.
• The models classify the risk of low back disorders (LBDs) due to lifting tasks.
• Our results are less optimistic than those reported in several other studies.
• We offer insights into how the risk factors that contribute to LBDs act.
• These insights could help to establish some guidelines to prevent injuries.

Low back disorders (LBDs) due to manual material lifting tasks have become a significant issue which affects the quality of life of industrial population of workers in the U.S. and has enormous economic impact. For the last three decades researchers have been trying to understand the phenomena of LBDs and develop practical guidelines which could prevent these injuries from happening or limit the severity of these injuries after they have already occurred. One of the research streams concentrated on creating and testing various classification models based on a landmark Marras data set. The goal of these models was to categorize manual lifting jobs as low risk or high risk with respect to LBDs. This paper summarizes and critiques the previous approaches as some of them yielded unrealistically high classification accuracy rates. The paper also proposes an adaptive neuro-fuzzy inference system (ANFIS) to classify tasks into high risk or low risk. To our best knowledge ANFIS has not been used in this context yet and has not been used for classification of a binary target variable. The paper also compares the classification performances of the different parameters or configurations of ANFIS. The ANFIS model appears to be a viable option for risk classification as it exhibits the classification accuracy rates consistent with several previous studies. More importantly ANFIS generates easy to interpret control surfaces, membership functions, and fuzzy rules, thus allowing one to get a deeper insight into the relationships between risk factors which interact with each other in a complex and nonlinear way. Such insights could prove to be very useful for the much needed efforts to better understand LBDs.