Expert System Based Diabetes Diagnoses Using Ensemble Model

Abstract

Abstract The title “Expert Systems” was given by John McCarthy, in which he entitled his study as “Applied Science and Technology” of making quick witted devices”. The development of ‘thinking’ computer systems popularly known as “Expert Systems” furthermore; also known as “Synthetic Intelligence” which explains the idea of learning these algorithms so that they can act intelligently. The mindset of software programs is basically to execute their activities intelligently like humans. The designs that allow software programs to work in such a way which frames its society as the ability to act intelligently known as Expert systems. The ability of these systems in order to use relevant connection between database which can be used in primary level screening for various disease, proper medication as well as estimating various results. In today’s growth following approach is being used in all useful domains which are beneficial for the society so that complex issues could be solved on time. This domain was being setup on the increasing demand the central attribute of human – mind can be imitated via these Expert systems.The work executed in this domain depicts how these expert systems could work intelligently provided if the data fed to this system is precisely accurate and tested by cross validation. To carry out this research total fourteen classifiers are being studied out of which we find four algorithms more suitable for this research namely Artificial Neural Network, K-Nearest Neighbor, Support vector machine, Naïve bayes. On top of it, to make our results more accurate and efficient, ensemble technique is used which predicts the output by taking the majority number of votes from the results predicted by these algorithms; in order to make this tool more efficient and accurate. The software tools which are being used are to execute this research are matrix laboratory and weka 3.6.13. After detailing of medical history, we prepared a rich dataset of around 400 people which is taken in the form of questionnaire from different sections of the society on the basis of ten physiological parameters. This enriched data has been tested and validated in terms of accuracy and prediction of correct data. This trained dataset is fed to graphical user interface and is tested so that it can predict correct output on the given input data provided to it.The values which are provided to the interface are five numeric values and rest are nominal values. The fig which illustrates the working of this tool is shown in Fig. 3. The foremost aim of this proposed work is to build an intelligent tool which is capable of predicting accurate results and provide correct data so that it could prove helpful in medical domain and can be used for initial screening of patients at an earlier stage so that proper treatment can be followed by them.Talking about the privacy of these algorithms out of all algorithms ANN outperformed with accuracy of 97.5% and our proposed ensembling technique assured the accuracy of 98%.