Authors: Mensah Kwabena Patrick
Addresses: Department of Computer Science, University for Development Studies, P.O. Box 24, Upper East, Ghana
Abstract: A simple search on Google with the phrase 'Mental Health in Ghana' will produce lots of hits including news items through publications to mental health bills. A lot of information is hidden in such unstructured data and could be mined to improve mental health in Ghana. For this reason, lexicon-based sentiment analysis and supervised machine learning was applied on a corpus of 28 journal articles and two news items in mental health. The sentiment score for the lexicon-based text analysis was negative indicating negative impression on mental health. SVM classification outperformed other algorithms with accuracy of 0.839. MAXENT was the worst performer (accuracy 0.774). F-score for SVM was 0.67; 0.64 for MAXENT and 0.59 for RF. 83.9% accuracy means that we can effectively predict human behaviour towards mental health via text mining because our behaviours are influenced by opinions we carry.
Keywords: mental health; classification; sentiment analysis; textual prediction; machine learning; support vector machines; SVM; regression forest; maximum entropy; Ghana; text mining; text analysis; behaviour prediction.
International Journal of Knowledge Engineering and Data Mining, 2015 Vol.3 No.3/4, pp.274 - 285
Received: 12 May 2015
Accepted: 13 Oct 2015
Published online: 05 Jan 2016 *