Mgini - Improved Decision Tree using Minority Class Sensitive Splitting Criterion for Imbalanced Data of Covid-19

In the time of COVID-19, medical facilities struggling to fight against the pandemic. Most of the countries face a tough time fighting against this virus outbreak. Even developed countries are struggling to deal with this virus outbreak. Common problem countries face is a lack of medical staff and medical equipment. Machine learning has the potential to play an important role in a different area of medical facilities. With the help of the machine learning model, an effective diagnostic tool can be built which helps in the time of scarcity of medical staff. However medical data is imbalanced and this skew nature of data prevent machine learning algorithm from achieving high accuracy. To deal with this problem of imbalanced data, we proposed a modified decision tree algorithm that uses a minority sensitive Gini index called Mgini. In an imbalanced dataset of COVID-19, it is important to focus on the reduction of overall misclassification cost instead of trying improvement in accuracy value. Mgini is useful splitting criteria when the misclassification cost of the minority sample is huge as compared to the majority class. The use of this proposed new Gini index as a splitting criterion in the decision tree reduces the misclassification cost. Mgini based decision tree has higher accuracy and low misclassification cost as compare to the traditional Gini index based CART algorithm. Our proposed cost-sensitive approach improves imbalanced data classification without the use of data level sampling techniques.