Misclassification Cost Sensitive Learning For Predicting Gonorrhea Infection Status In Ghana

Abstract

Gonorrhea, which is one of the most frequently reported sexually transmitted infection is caused by a bacterium called Neisseria gonorrhoeae. This disease can causes a serious public health problem worldwide, with about 88 million new infections occurring each year. Failure to treat this disease can result into pelvic inflammatory disease (PID), chronic pain and also damage the female reproductive organ. In males it can lead to reduced fertility and sterility. In developing countries, the unavailability of diagnostic capacity due to cost, lack of equipment and trained personnel has led to the syndrome based management of sexually transmitted infection (STI). Due to these challenges, there is the need for statistical models for gonorrhoea diagnosis which can easily be obtained and implemented with the appropriate expertise. In diagnosing sexually transmitted infection, a false positive has different impact than vice versa. Assuming equal misclassification cost in such models can lead to incorrect decisions and also incur financial cost and harm to the patient. Many classifiers do not allow integration of cost into model development process but rather are designed to improve prediction accuracy assuming equal misclassification cost. The aim of the study is to develop cost sensitive statistical models for predicting gonorrhoea infection. For the data used for the study, 80% was used for training and 20% for testing. The results indicated that, the cost sensitive classifiers had a reduced total classification cost than the cost insensitive classifiers. Also, the classification cost of all laboratory diagnostic method except culture was lower than the cost sensitive and insensitive model. The class distribution weakly affected the cost sensitive classifiers but not the cost insensitive classifiers.