Journal of Scientific Research and Reports, ISSN: 2320-0227,Vol.: 3, Issue.: 23 (01-15 December)
Application of Data Mining Techniques to Audiometric Data among Professionals in India
J. Majumder1 and L. K. Sharma1* 1National Institute of Occupational Health (ICMR), Ahmedabad-380016, India.
J. Majumder1 and L. K. Sharma1*
1National Institute of Occupational Health (ICMR), Ahmedabad-380016, India.
(1) Prof. William Ebomoyi, Department of Health Studies, College of Health Sciences, Chicago State University, USA.
Complete Peer review History: http://www.sciencedomain.org/review-history/6134
Aims: Noise induced hearing loss (NIHL) is among the principal occupational health hazard. To illustrate that, in order to enrich the database on audiometric status and fast dissemination of knowledgebase, data mining techniques are imperative tools.
Study Design: A cross sectional study design was used.
Place and Duration of Study: Pure tone audiometric data of both ears of drivers that have 10 years working experience and office workers from Kolkata City, India were recorded.
Methodology: The data were subjected to both unsupervised and supervised learning techniques, in turn, in order to train the classifier that determines the clusters for newly generated cases. Expectation Maximization (EM), k-means, Linear Vector Quantization (LVQ), and Self Organization Map (SOM) unsupervised learning techniques were utilized.
Results: Silhouette Plot (SP) validation showed that 93.3% of the considered cases for the left ear and 85.8% for the right ear were correctly classified. These metadata were further subjected to supervised learning algorithm to achieve a high level correctly classified result, in which, each cluster bears its class label. Naïve Bays Classifier (NBC) recorded, as accurate (98.8%) for both left and right ears. The high accuracy of supervised learning algorithms, cross validated with 10-fold cross validation tends to predict the class of audiometric data whenever a newly generated data are introduced.
Conclusion: This feasibility of using machine learning and data classification models on the audiometric data would be an effective tool in the hearing conservation program for individuals exposed to noisy environments in their respective workplaces.
Hearing threshold; cluster analysis; unsupervised learning; supervised learning; cross validation.
DOI : 10.9734/JSRR/2014/12700Review History Comments