Project Performance Evaluation Using Deep Belief Networks

カテゴリ: 論文誌(論文単位)

グループ名: 【C】電子・情報・システム部門

発行日: 2012/02/01

タイトル(英語): Project Performance Evaluation Using Deep Belief Networks

著者名: Alick Nguvulu (Graduate School of Information Science and Technology, Hokkaido University), Shoso Yamato (Graduate School of Systems and Information Engineering, Tsukuba University), Toshihisa Honma (Graduate School of Information Science and Technology,

著者名(英語): Alick Nguvulu (Graduate School of Information Science and Technology, Hokkaido University), Shoso Yamato (Graduate School of Systems and Information Engineering, Tsukuba University), Toshihisa Honma (Graduate School of Information Science and Technology, Hokkaido University)

キーワード: Deep Belief Networks，Machine Learning，Neural Networks，Project Management，Project Performance

要約(英語): A Project Assessment Indicator (PAI) Model has recently been applied to evaluate monthly project performance based on 15 project elements derived from the project management (PM) knowledge areas. While the PAI Model comprehensively evaluates project performance, it lacks objectivity and universality. It lacks objectivity because experts assign model weights intuitively based on their PM skills and experience. It lacks universality because the allocation of ceiling scores to project elements is done ad hoc based on the empirical rule without taking into account the interactions between the project elements. This study overcomes these limitations by applying a DBN approach where the model automatically assigns weights and allocates ceiling scores to the project elements based on the DBN weights which capture the interaction between the project elements. We train our DBN on 5 IT projects of 12 months duration and test it on 8 IT projects with less than 12 months duration. We completely eliminate the manual assigning of weights and compute ceiling scores of project elements based on DBN weights. Our trained DBN evaluates monthly project performance of the 8 test projects based on the 15 project elements to within a monthly relative error margin of between ±1.03 and ±3.30%.

本誌: 電気学会論文誌C（電子・情報・システム部門誌） Vol.132 No.2 （2012） 特集：多様な情報社会に適応するシステム技術

本誌掲載ページ: 306-312 p

原稿種別: 論文／英語

電子版へのリンク: https://www.jstage.jst.go.jp/article/ieejeiss/132/2/132_2_306/_article/-char/ja/