Asian Research Thesis Index

Abstract

In this era of information and technology data mining has gained much fame. Millions of
versatile data records in various forms such as text, digits and images are going to store in
databases and online data repositories. Machine learning techniques are playing vital role in
analyzing such bulk of data in better way. Health department is considered as one of the most
significant domain of generating huge collection of data associated to patient?s care, diagnostics,
analysis and recommendations in various contexts based on disease and medical situations. The
analysis of health care data can be very helpful for diagnosis of patients and decision making. A
number of comparative researches in machine learning techniques have been performed in the
literature on health data; however most of these approaches have been limited to a single dataset
analysis, focused on a small number of parameters evaluation such as accuracy measurement and
lack of graphical representation of statistical performance metrics. There is need to use more
parameters and multiple data sets in order to evaluate machine learning algorithms for their
maximum performance. The purpose of this research work was to propose and conduct empirical
analysis of multiple machine learning classifiers through accuracy, precision, sensitivity,
specificity and F-measure parameters to measure their maximum performance on health data. In
this regard Diabetes, Kidney, Liver, Lungs and Heart datasets have been analyzed using Na?ve
Bayes, LMT, SMO, JRip and J48 Decision Tree classifiers. It has been concluded from analysis
that J48 classifier has shown optimal functionality on health datasets having large number of
attributes. It has shown high accuracy and F-measure value on CKD (Chronic Kidney Dataset)
dataset that is the highest ratio among other classifiers. While in case of small datasets (Lung
cancer) Na?ve Bayes and SMO has beaten other classifiers. In graphical representation ROC
curve has proved that Na?ve Bayes classifiers presented maximum performance. Precision-Recall
curve proved that J48 has beaten other classifiers. Graphical representation of the results of
different statistical performance metrics of machine learning Algorithms have also been
provided.

Add/Update Thesis