A Precise SVM Classification Model for Predictions with Missing Data

In a well-studied and controlled research work, missing data are the common occurrence can have the significant influence on the results and accuracy of the study. Missing data can cause biased estimates and leads to the wrong conclusions. This paper develops a model to classify a two-class problem and report the classification accuracy over the stratified 10 folds cross-validation based on the provided data with missing values. The dataset contains 14 features and two classes, in which there are missing data without any expressions with an arbitrary pattern. To classify the data and predicate the missing values, the data is preprocessed in the first step, and it has been sent to the proposed SVM (Support Vector Machine) model for further processes. In order to improve the accuracy of classification, the metaheuristic methods such as GSM (Grid Search Method), PSO (Particle Swarm Optimization), and GA (Genetic Algorithm) have been used to extract the best parameters, C (penalty parameter) and g (kernel function parameter) of the SVM, and then, their F-Measures have been calculated to choose the best model