Analyzing Children’s Data Using Machine Learning: A Case Study in Ethiopia

Abstract

This research focuses on classification of children into four classes namely orphan, single orphan, vulnerable and safe. The aim of this classification is to help the outside donors of the Love for Children Organization and further to get full information about each child for internal purpose of the organization. To achieve this three classification techniques were used which are Decision tree, Bayesian learning and Neural network within the framework of KDD (Knowledge Discovery in Databases) data mining model. The children dataset was collected, cleaned, transformed and integrated for experimenting with the classification model. The final dataset consists of 17044 records that have been experimented and evaluated against their performances. The collection of dataset was experimented with the 10-fold cross-validation and splitting the datasets in to 70/30%, 66/44% and 50/50% for training/and for testing respectively. Additionally, a comparison of decision tree (98.83 %,), bayesian learning (98.32%) and neural network (98.86%) model in terms of the overall classification accuracy and their advantage was made. The research concludes that decision tree (98.83%) should be selected as a model because it gives better results than Bayesian learning (98.32%) and better advantage over Neural Network (98.86) for the classification of organizations children.