期刊名称：International Journal of Computer Science and Network Security
印刷版ISSN：1738-7906
出版年度：2018
卷号：18
期号：10
页码：29-35
出版社：International Journal of Computer Science and Network Security
摘要：The aim of research work is towards the prevalence of diabetes mellitus to improve classification accuracy and cost/benefit predictions on real-life dataset. This paper aims the real-life diabetic patients from the country ‘Nigeria’ (an African state) to test the experiment.Our proposed methodology consist of two parts weka classification for the accuracy measurement ratios by applying 150 machine learning classifiers and compare the results with previous work, secondly’ the improvement in clustering with positive and negative instances ratios in terms of initial care predictions by cost benefit analysis.For our Diabetes dataset of 281 instances with 100 attributes, auto-weka has performed 2,440 evaluations with the error rate of 0.014 % by the time limit of 120 minutes. We have achieved the accuracy ratio of 99.64% as compared to 92.6% by “AdaboostM1” classification in mean time of 0.015 seconds to build the model and properly classified instance ratio of 86% by improving k-means.One of the benefits of our proposed model is that it avoids deleting the original data, which ensures the high quality in experiments. And the other benefit is, model can be applied to the other datasets to attain the best accuracy and cost analysis results.
