Evaluation of machine learning classification algorithms & missing data imputation techniques

In this work, we present a performance comparison of the Multi Layer Perceptron (MLP), Support Vector Machines (SVM) and Voted Perceptron (VP) when applied to a social signal processing task. The signal processing task is in the field of computational politics where the aim is to predict the political parties of American congress members based on their response to certain questions. Using this dataset which is publicly available, we investigate the use of four methods to impute or approximate missing values. The four imputed datasets are used to train MLP, SVM and VP classifiers to associate the congress members' responses to their political party affiliation and we compare the results from the three classifiers. The aim is to design a practical system or model to be able to predict another person's political affiliations based on their responses to similar questions. The obtained experimental results suggest that machine learning classifiers can be used to accurately predict an individual's political leaning.