Missing values are most likely to be present in voluminous datasets that often lead to poor performance of the decision-making system. The present work carries out an experiment with a crime dataset that deals with the existence of missing values in it. The proposed methodology depicts a graph-based approach for selecting important features relevant to crime after estimating the missing values with the help of a multiple regression model. The method selects some features with missing values as important features. The selected features subsequently undergo some classification techniques that help in determining the importance of missing value estimation without discarding the feature for crime analysis. The proposed method is compared with existing feature selection algorithms and it promises a better classification accuracy, which shows the importance of the method.