Optimised feature selection and categorisation of medical records with multi kernel boosted support vector machine

With the fast growth of internet and mobile usage, huge volumes of medical documents, which contain information of patients, diagnostic, past disease history and medication, are being generated electronically. In the field of text mining, document categorisation has become one of the emerging techniques due to large volume of documents in the form of digital data. The main objective of the proposed work is to identify disease treatment relationships and predict the diseases among medical articles. In this paper, highly relevant and more correlated features have been extracted using probabilistic latent Dirichlet allocation (P-LDA) and randomised iterative feature selection approach. These features were classified with multi kernel boosted support vector machine (MKB-SVM) and then their performance was evaluated on both PubMed and MEDLINE databases. Performance evaluation of the proposed approach on DB-1 and DB-2 was 98.7% and 92%, respectively. The evaluation illustrated that the proposed approach outperformed the existing state-of-the-art classification methods.