DEVELOPMENT CB ALGORITHM TO CONSTRUCT BAYESIAN NETWORK STRUCTURE FROM INCOMPLETE DATA.

• Main Author: Tommy Argo Simanjuntak(NIM : 13505603), Humasak
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/8159
• Institution: Institut Teknologi Bandung

Abstract: <br /> <br /> <br /> <br /> <br /> <br /> Data mining is the core process of Knowledge Discovery in Databases (KDD) that mine pattern or knowledge from voluminous data. One of representation model that can be used for data mining is Bayesian Network (BN). BN consists of network structure called Directed Acyclic Graph (DAG) that represents conditional independency and network parameter, represents Joint Probability Distribution (JPD). <br /> <br /> <br /> <br /> <br /> <br /> There exists two approaches to construct BN structure, search & scoring and dependency analysis approaches. CB algorithm is a hybrid algorithm that combines search & scoring and dependency analysis approaches to construct BN structure from data. The weakness of CB algorithm is construction of BN structure cannot be done from incomplete data. This matter occurs because some step of CB algorithm (step 2, 6, and 8) requiring complete data. Therefore, in this final project, CB algorithm that developed to construct structure from incomplete data was analysized, implemented, and tested to get the best algorithm. There are two type of CB algorithm will developed, they are: CB algorithm with ignore tuple method and CB algorithm (CB algorithm that combine with BSEM algorithm and ignore tuple method). <br /> <br /> <br /> <br /> <br /> <br /> By doing analysis, we conclude that CB algorithm better than CB algorithm with ignore tuple method. But, time performance of CB algorithm longer than CB algorithm with ignore tuple method. <br /> <br /> <br /> <br /> <br /> <br /> Next task is proving the analysis result, by designing and implementing CB algorithms software for incomplete data, then test it with cases study called Visit to Asia and Fire network. Software is developed using Delphi 7.0 which runs in windows platform. Test objective is to observe best development of CB algorithm to construct BN structure from incomplete data. <br /> <br /> <br /> <br /> <br /> <br /> This final projects conclusion is CB algorithm better than CB algorithm with ignore tuple method to construct BN structure from incomplete data. <br />