Imputation of Missing Value Using Dynamic Bayesian Network for Multivariate Time Series Data

• Main Author: Pauli Susanti , Steffi
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/24381
• Institution: Institut Teknologi Bandung

As the decision making process growing more complex, time series data and <br /> <br /> multivariate is necessary to accommodate existing needs. Time series data is <br /> <br /> important data in management, planning, and decision making. To accommodate <br /> <br /> these needs, data is processed using data mining to see trends that occur in the data <br /> <br /> so that it can be used for consideration in decision making. Unfortunately, when we <br /> <br /> try to get information from data there are many problems encountered, one of them <br /> <br /> is missing value. Missing value can causes the results of data processing inaccurate <br /> <br /> with real circumstances. <br /> <br /> Imputation method can be used to handle missing values. Before imputation <br /> <br /> process, data is modeled with Dynamic Bayesian Network (DBN). DBN is selected <br /> <br /> because it can maintain the relationship between attributes in the data. After data <br /> <br /> is modeled into DBN, then imputation is done using prediction approach. Generally <br /> <br /> predictions are used to guess the data at the end of the dataset, but in this study the <br /> <br /> prediction method is used to fill the missing values of the dataset. The algorithm <br /> <br /> used for the imputation is Support Vector Regression (SVR). SVR was chosen <br /> <br /> because of its good performance compared to other algorithms. <br /> <br /> Testing is done by using Symmetric Mean Absolute Percentage Error (SMAPE). As <br /> <br /> a result, the use of the DBN model proved to improve the accuracy of predicted <br /> <br /> values for imputation performed with SVR.