An evaluation of classification algorithms for manual material handling tasks based on data obtained using wearable technologies

With recent progress in wearable measurement systems, physical exposures can be feasibly assessed at high precision in the workplace. Such systems, however, generally lack contextual information for a given job (e.g., task type, duration). To extract such information, we explored three classification algorithms to classify manual material handling (MMH) tasks during a simulated job in a laboratory, using several combinations of outputs from commercially available inertial motion capture and in-shoe pressure measurement systems. A total of 10 participants completed three replications of four cycles of a simulated job. Precision and recall values of ≥ ∼90% and 80%, respectively, and errors in estimated task duration of < ∼14%, could be achieved across the MMH task examined. Classification performance, however, varied between classification algorithms, input data sets and task types. Overall, combining wearable technology with task classification could be an effective approach for field-based exposure assessment, though field-testing is needed to demonstrate the applicability of this method.

Combining wearable technologies with task classification was explored to extract exposure context, specifically task type and duration. Results supported that task classification can facilitate the use of wearable technologies in field-based exposure assessment, specifically by aiding in task identification from within the rather large data sets obtained from these technologies.