Abstract
| Original language | English |
|---|---|
| Title of host publication | Proceedings of the 29th ICOH, International Congress on Occupational Health. Cape Town, South Africa, 22 - 27 March 2009 |
| Pages | 539 |
| Publication status | Published - 2009 |
| MoE publication type | Not Eligible |
| Event | 29th ICOH, International Congress on Occupational Health - Cape Town, South Africa Duration: 22 Mar 2009 → 27 Mar 2009 |
Conference
| Conference | 29th ICOH, International Congress on Occupational Health |
|---|---|
| Country | South Africa |
| City | Cape Town |
| Period | 22/03/09 → 27/03/09 |
Fingerprint
Keywords
- personnel screening
- well-being
- neural networks
Cite this
}
A neural network-based method for automated assessment of well-being based on personnel screening questionnaires. / van Gils, Mark; Suovanen, Jukka; Merilahti, Juho; Pärkkä, Juha; Rautamo, Henri.
Proceedings of the 29th ICOH, International Congress on Occupational Health. Cape Town, South Africa, 22 - 27 March 2009. 2009. p. 539.Research output: Chapter in Book/Report/Conference proceeding › Conference abstract in proceedings › Scientific
TY - CHAP
T1 - A neural network-based method for automated assessment of well-being based on personnel screening questionnaires
AU - van Gils, Mark
AU - Suovanen, Jukka
AU - Merilahti, Juho
AU - Pärkkä, Juha
AU - Rautamo, Henri
N1 - Project code: 22275
PY - 2009
Y1 - 2009
N2 - There is a need for objective, simple methods to assess employees' well-being. Currently, a healthcare professional assesses this partly using experience-based knowledge, making it difficult to use explicit rules to automate this task. We aimed to develop classifiers learning from data from questions in personnel screening questionnaires. Our goals were: a) to define an efficient subset of questions and b) to develop a classifier for well-being grading. Screening data from occupational health checks in a random selection of the working population in various parts of Finland were used. Data contained 98 answers on medication usage, pain experience, psychological factors, stress, lifestyle etc. A healthcare professional scored well-being on a scale from 1 to 3 (indicating seriousness in decrease of well-being). 1063 subjects were used. For a), relationships between variables were assessed using correspondence analyses. For b), linear and non-linear regression and neural networks (backpropagation networks) were used with as input the variables found from a) and as desired output the well-being scores. Final performance was assessed using an independent test set of 89 subjects. Significant correspondences between different questions were found. A subset of 9 independent questions proved to be efficient. These questions relate to weight, dizziness, sports activities, pain experience, views on life, and personal assessment of the ability to continue work. A neural network gave the best results with an accuracy of 83% on the test set, and sensitivity 85% and specificity 83% for separating 'reduced' from 'no reduced' well-being. It is possible to construct a classifier for severity of decrease in well-being that is in good agreement with a healthcare professional's opinion. That performance can be obtained with a relatively small number of questions. This allows implementation of a classifier in quick-to-complete questionnaires that can be used routinely.
AB - There is a need for objective, simple methods to assess employees' well-being. Currently, a healthcare professional assesses this partly using experience-based knowledge, making it difficult to use explicit rules to automate this task. We aimed to develop classifiers learning from data from questions in personnel screening questionnaires. Our goals were: a) to define an efficient subset of questions and b) to develop a classifier for well-being grading. Screening data from occupational health checks in a random selection of the working population in various parts of Finland were used. Data contained 98 answers on medication usage, pain experience, psychological factors, stress, lifestyle etc. A healthcare professional scored well-being on a scale from 1 to 3 (indicating seriousness in decrease of well-being). 1063 subjects were used. For a), relationships between variables were assessed using correspondence analyses. For b), linear and non-linear regression and neural networks (backpropagation networks) were used with as input the variables found from a) and as desired output the well-being scores. Final performance was assessed using an independent test set of 89 subjects. Significant correspondences between different questions were found. A subset of 9 independent questions proved to be efficient. These questions relate to weight, dizziness, sports activities, pain experience, views on life, and personal assessment of the ability to continue work. A neural network gave the best results with an accuracy of 83% on the test set, and sensitivity 85% and specificity 83% for separating 'reduced' from 'no reduced' well-being. It is possible to construct a classifier for severity of decrease in well-being that is in good agreement with a healthcare professional's opinion. That performance can be obtained with a relatively small number of questions. This allows implementation of a classifier in quick-to-complete questionnaires that can be used routinely.
KW - personnel screening
KW - well-being
KW - neural networks
M3 - Conference abstract in proceedings
SP - 539
BT - Proceedings of the 29th ICOH, International Congress on Occupational Health. Cape Town, South Africa, 22 - 27 March 2009
ER -