A Study of the Unbalanced Image Dataset for Classification Problem

In the case of experimental data it is evident that the dataset obtained is unbalanced. The main reason for the same is the probability of the occurrence of a particular event in real life. As these events have very less probabilities hence the collection of such sample data is very difficult. Therefore, various approaches have been suggested to overcome this problem. This paper compares the two approaches for handling the unbalanced data. In the first approach, the minimum class with random selection was used to balance the data and in the second approach, the maximum class with random duplicate is used to balance the data. Based on the results of the sugarcane disease detection using densenet with SVM it was evident that incorporating the maximum data available is helping the machine learning model to more effectively.