Imputation of missing value using dynamic Bayesian network for multivariate time series data

Abstract

Time series and multivariate data are required to accommodate more complex decision making. Data are processed using data mining techniques in order to obtain valuable trends in the data that can be used to support in decision making processes. Unfortunately, we often encounter a lot of problems in preparing the data for data mining process. One of the problem is missing values. Missing values in data may causes inaccurate results of data processing. Imputation are used to handle missing values. In this thesis missing value are handled using Dynamic Bayesian Network (DBN). DBN is a useful technique to maintain the relationships between attributes of data. The results of the prediction are used to fill in the missing values in the data. Support Vector Regression (SVR) algorithm is used for predicting the missing values. It is chosen for its good performance in comparison to other similar algorithms. Validation of the technique is carried out by using Symmetric Mean Absolute Percentage Error (SMAPE). SMAPE used to count an error rate for prediction model. The use of the DBN of feature selection for SVR can't decrease the error rate of the model.