Statistical and Neural Integration of Data and Text Mining for Supporting Audiology Knowledge Extraction

We are working on the data mining of audiology patient records, looking for factors influencing which patients would most benefit from being fitted with a hearing aid. Various research groups have used statistics and neural networks for integration of data, such as the chi-squared test or self organizing maps (SOMs). This motivates our research into a new architecture which combines neural network and statistical techniques suitable for mining heterogeneous audiology records. Audiology records contain the following three different types of data: 1. Audiograms (graphs of hearing ability at different frequencies) 2. Structured tabular data (gender, date of birth, etc) 3.Unstructured text (specific observations made about each patient in a field for comments/remarks) So far, we have performed clustering of hearing aid patient audiograms with an SOM, and clustering of audiograms using the k-means algorithm to identify 3 main groups of hearing aid wearers.