Gordon Ondego | JOMO KENYATAA UNIVERSITY OF AGRICULTURE AND TECHNOLOGY (original) (raw)

Uploads

Thesis Chapters by Gordon Ondego

Research paper thumbnail of A Comparative Analysis of Machine Learning Algorithms to Classify Cause of Death from Verbal Autopsy Data

With an increased effort to reduce mortality rate in most developing countries, accurate informat... more With an increased effort to reduce mortality rate in most developing countries, accurate information on the causes of such mortalities is a very crucial component for the development and formulation of health policy, strategies and other key critical decisions in the health
sector. However there is lack of complete, accurate and reliable vital registration system that is expected to generate and report accurate causes of death information for health intervention policies and other programs. This research sets out to make a comparative evaluation of two most common supervised machine learning approaches Naive Bayes (NB) and J48 decision tree which builds a decision tree in the context and with the aid of Institute for Health Metrics
and Evaluation (IHME) Verbal Autopsy (VA) dataset.
This research also focuses on experimental comparison of these two state of art supervised learning techniques with respect to their accuracy of correctly classified instances, incorrectly classified instances and very important Receiver Operating Characteric (ROC)
Area which helps in understanding the classification model and their results, which can also help other researchers in making decision for the selection in classification model based on their data and number of attributes.
With reference from several conference papers published recently, journals and other resources, the research was accomplished by training and testing the selected algorithms with the same datasets using a 10 fold cross validation method in Waikato Environment for Knowledge Analysis (WEKA) platform. The experiments carried out in this research are about classification accuracy, sensitivity and specificity using true positive (TP) and false positive (FP) in confusion matrix generated by the respective algorithms. The results obtained shows that J 48 decision tree algorithms out performs Naïve Bayes in terms of accuracy, recall, precision and F score. The perfection of these algorithms in the classification task is further explained with the analysis of ROC curve. The results obtained from the study indicate that J48 decision tree algorithm performs better than the Naïve Bayes classifier. A prototype
has been developed based on the J48 decision tree algorithm because it exhibits good performance in the prediction of cause of death from the verbal autopsy data set.
This prototype can be used by medical experts both in the private and public hospitals to make more timely and consistent diagnosis of the causes of death from the verbal autopsy for those deaths occuring outside health institutions.

Papers by Gordon Ondego

Research paper thumbnail of Comparative Analysis of Machine Learning Algorithms for Mycobacterium Tuberculosis Protein Sequences on the Basis of Physicochemical Parameters

Journal of Medical Imaging and Health Informatics, 2014

Research paper thumbnail of A Machine learning Approach to predict cause of death from Verbal Autopsy Data

Research paper thumbnail of A comparative study of decision Tree and Naïve Bayesian Classifiers on Verbal Autopsy Datasets

With an increased effort to reduce mortality rate in most developing countries, accurate informat... more With an increased effort to reduce mortality rate in most developing countries, accurate information on the causes of such mortalities is a very crucial component for the development and formulation of health policy, strategies and other key critical decisions in the health sector. However there is lack of complete, accurate and reliable vital registration system that is expected to generate and report accurate causes of death information for health intervention policies and other programs. This research sets out to make a comparative evaluation of two most common supervised machine learning approaches Naive Bayes (NB) and J48 decision tree which builds a decision tree in the context and with the aid of Institute for Health Metrics and Evaluation (IHME) Verbal Autopsy (VA) dataset. This research also focuses on experimental comparison of these two state of art supervised learning techniquues with respect to their accuracy of correctly classified instances, incorrectly classified instances and very important Receiver Operating Characteric (ROC) Area which helps in understanding the classification model and their results, which can also help other researchers in making decision for the selection in classification model based on their data and number of attributes. With reference from several conference papers published recently, journals and other resources, the research was accomplished by training and testing the selected algorithms with the same datasets using a 10 fold cross validation method in Waikato Environment for Knowledge Analysis (WEKA) platform. The experiments carried out in this research are about classification accuracy, sensitivity and specificity using true positive (TP) and false positive (FP) in confusion matrix generated by the respective algorithms. The results obtained shows that J 48 decision tree algorithms out performs Naïve Bayes in terms of accuracy, recall, precision and F score. The perfection of these algorithms in the classification task is further explained with the analysis of ROC curve. The results obtained from the study indicate that J48 decision tree algorithm performs better than the Naïve Bayes classifier. A prototype has been developed based on the J48 decision tree algorithm because it exhibits good performance in the prediction of cause of death from the verbal autopsy data set.This prototype can be used by medical experts both in the private and public hospitals to make more timely and consistent diagnosis of the causes of death from the verbal autopsy for those deaths occuring outside health institutions. v TABLE OF CONTENTS

Research paper thumbnail of A Comparative Analysis of Machine Learning Algorithms to Classify Cause of Death from Verbal Autopsy Data

With an increased effort to reduce mortality rate in most developing countries, accurate informat... more With an increased effort to reduce mortality rate in most developing countries, accurate information on the causes of such mortalities is a very crucial component for the development and formulation of health policy, strategies and other key critical decisions in the health
sector. However there is lack of complete, accurate and reliable vital registration system that is expected to generate and report accurate causes of death information for health intervention policies and other programs. This research sets out to make a comparative evaluation of two most common supervised machine learning approaches Naive Bayes (NB) and J48 decision tree which builds a decision tree in the context and with the aid of Institute for Health Metrics
and Evaluation (IHME) Verbal Autopsy (VA) dataset.
This research also focuses on experimental comparison of these two state of art supervised learning techniques with respect to their accuracy of correctly classified instances, incorrectly classified instances and very important Receiver Operating Characteric (ROC)
Area which helps in understanding the classification model and their results, which can also help other researchers in making decision for the selection in classification model based on their data and number of attributes.
With reference from several conference papers published recently, journals and other resources, the research was accomplished by training and testing the selected algorithms with the same datasets using a 10 fold cross validation method in Waikato Environment for Knowledge Analysis (WEKA) platform. The experiments carried out in this research are about classification accuracy, sensitivity and specificity using true positive (TP) and false positive (FP) in confusion matrix generated by the respective algorithms. The results obtained shows that J 48 decision tree algorithms out performs Naïve Bayes in terms of accuracy, recall, precision and F score. The perfection of these algorithms in the classification task is further explained with the analysis of ROC curve. The results obtained from the study indicate that J48 decision tree algorithm performs better than the Naïve Bayes classifier. A prototype
has been developed based on the J48 decision tree algorithm because it exhibits good performance in the prediction of cause of death from the verbal autopsy data set.
This prototype can be used by medical experts both in the private and public hospitals to make more timely and consistent diagnosis of the causes of death from the verbal autopsy for those deaths occuring outside health institutions.

Research paper thumbnail of Comparative Analysis of Machine Learning Algorithms for Mycobacterium Tuberculosis Protein Sequences on the Basis of Physicochemical Parameters

Journal of Medical Imaging and Health Informatics, 2014

Research paper thumbnail of A Machine learning Approach to predict cause of death from Verbal Autopsy Data

Research paper thumbnail of A comparative study of decision Tree and Naïve Bayesian Classifiers on Verbal Autopsy Datasets

With an increased effort to reduce mortality rate in most developing countries, accurate informat... more With an increased effort to reduce mortality rate in most developing countries, accurate information on the causes of such mortalities is a very crucial component for the development and formulation of health policy, strategies and other key critical decisions in the health sector. However there is lack of complete, accurate and reliable vital registration system that is expected to generate and report accurate causes of death information for health intervention policies and other programs. This research sets out to make a comparative evaluation of two most common supervised machine learning approaches Naive Bayes (NB) and J48 decision tree which builds a decision tree in the context and with the aid of Institute for Health Metrics and Evaluation (IHME) Verbal Autopsy (VA) dataset. This research also focuses on experimental comparison of these two state of art supervised learning techniquues with respect to their accuracy of correctly classified instances, incorrectly classified instances and very important Receiver Operating Characteric (ROC) Area which helps in understanding the classification model and their results, which can also help other researchers in making decision for the selection in classification model based on their data and number of attributes. With reference from several conference papers published recently, journals and other resources, the research was accomplished by training and testing the selected algorithms with the same datasets using a 10 fold cross validation method in Waikato Environment for Knowledge Analysis (WEKA) platform. The experiments carried out in this research are about classification accuracy, sensitivity and specificity using true positive (TP) and false positive (FP) in confusion matrix generated by the respective algorithms. The results obtained shows that J 48 decision tree algorithms out performs Naïve Bayes in terms of accuracy, recall, precision and F score. The perfection of these algorithms in the classification task is further explained with the analysis of ROC curve. The results obtained from the study indicate that J48 decision tree algorithm performs better than the Naïve Bayes classifier. A prototype has been developed based on the J48 decision tree algorithm because it exhibits good performance in the prediction of cause of death from the verbal autopsy data set.This prototype can be used by medical experts both in the private and public hospitals to make more timely and consistent diagnosis of the causes of death from the verbal autopsy for those deaths occuring outside health institutions. v TABLE OF CONTENTS