A Novel Approach for Dealing with Missing Values in Machine Learning Datasets with Discrete Values

One of the problems that faces machine learning researchers is the incomplete datasets with missing values, knowing that most machine learning algorithms deal with complete datasets. ILA is one of these algorithms which deal only with datasets with complete discrete values. In this paper, a novel approach for dealing with missing values has been developed and tailored with ILA where the treatment of missing values is performed during the induction process. The proposed system is called ILA4. ILA4 has been tested on several datasets with different percentages of missing values. Its results also compared with some common methods for treating missing values. The results show that most of the time, the results ILA4 appear to be comparable to the best cases of some other well-known methods for dealing with missing values problem, namely; the most common value, the most common value restricted to a concept, and delete strategy.