Organization
Faculty of Business Data Science Professor
Title
Professor
External link
TAKAI,Keiji
Degree
-
博士 ( 2009.3 )
Research Areas
-
Natural Science / Applied mathematics and statistics
Education
-
Osaka University Graduate School, Division of Engineering Science
- 2009
-
Osaka University Graduate School, Division of Engineering Science
- 2005
-
Osaka University Faculty of Human Science
- 2003
-
Osaka University Graduate School, Division of Engineering Science
2009
-
Osaka University Graduate School, Division of Engineering Science
2005
Professional Memberships
-
The Behaviormetric Society of Japan
-
Japan Statistical Society
-
Japan Association for Consumer Studies
-
KES international
-
Psychometric Society
Papers
-
Model Selection with Missing Data Embedded in Missing-at-Random Data Reviewed
Keiji Takai, Kenichi Hayashi
Stats 6 ( 2 ) 495 - 505 2023.4
-
Incomplete-data Fisher scoring method with steplength adjustment Reviewed
Keiji Takai
Statistics and Computing 30 ( 4 ) 871 - 886 2020.2
-
A Framework of ASP for Shopping Path Analysis Reviewed
Katsutoshi Yada, Kei Miyazaki, Keiji Takai, Kohei Ichikawa
Proceedings - 2017 4th Asia-Pacific World Congress on Computer Science and Engineering, APWC on CSE 2017 49 - 54 2018.10
-
On the use of the selection matrix in the maximum like- lihood estimation of normal distribution models with missing data Reviewed
TAKAI,Keiji
Communications in Statistics – Theory and Methods 47,14,3992-3407 2018
-
Global Distribution of Watches A network analysis of trade relations Reviewed
Pierre-Yves Donze, Ken Ishibashi, Bo Wu, Yuta Kaneko, Kei Miyazaki, Keiji Takai
2017 17TH IEEE INTERNATIONAL CONFERENCE ON DATA MINING WORKSHOPS (ICDMW 2017) 605 - 611 2017
-
Finite-sample analysis of impacts of unlabeled data and their labeling mechanisms in linear discriminant analysis Reviewed
Kenichi Hayashi, Keiji Takai
COMMUNICATIONS IN STATISTICS-SIMULATION AND COMPUTATION 46 ( 1 ) 184 - 203 2017
-
Shop area visit ratio, stay time, and sales outcomes: In-depth analysis based on RFID data Reviewed
Zhen Li, Ken Ishibashi, Keiji Takai, Katsutoshi Yada
2015 2nd Asia-Pacific World Congress on Computer Science and Engineering, APWC on CSE 2015 2016.5
-
Effects of unlabeled data on classification error in normal discriminant analysis Reviewed
Keiji Takai, Kenichi Hayashi
JOURNAL OF STATISTICAL PLANNING AND INFERENCE 147 66 - 83 2014.4
-
A framework for analysis of the effect of time on shopping behavior
Keiji Takai, Katsutoshi Yada
JOURNAL OF INTELLIGENT INFORMATION SYSTEMS 41 ( 1 ) 91 - 107 2013.8
-
Asymptotic inference with incomplete data Reviewed
Keiji Takai, Yutaka Kano
Communications in Statistics: Theory and Methods 2012
-
A framework for analysis of the effect of time on shopping behavior Reviewed
高井啓二, 矢田勝俊
Journal of Intelligent Information Systems 41 2012
-
Exploration of Dependencies among Sections in a Supermarket Using a Tree-Structured Undirected Graphical Model Reviewed
Keiji Takai
12TH IEEE INTERNATIONAL CONFERENCE ON DATA MINING WORKSHOPS (ICDMW 2012) 324 - 331 2012
-
Analysis of NMAR missing data without specifying missing-data mechanisms in a linear latent variate model Reviewed
Yutaka Kano, Keiji Takai
JOURNAL OF MULTIVARIATE ANALYSIS 102 ( 9 ) 1241 - 1255 2011.10
-
Clockwise and anti-clockwise directions of customer orientation in a supermarket: Evidence from RFID data
Marina Kholod, Keiji Takai, Katsutoshi Yada
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 6883 ( 3 ) 304 - 309 2011
-
Finding Latent Groups of Customers via the Poisson Mixture Regression Model
Keiji Takai, Katsutoshi Yada
2011 IEEE INTERNATIONAL CONFERENCE ON SYSTEMS, MAN, AND CYBERNETICS (SMC) 3603 - 3608 2011
-
Constrained EM algorithm with projection method Reviewed
Keiji Takai
Computational Statistics 1-14 2011
-
Relation between Stay-Time and Purchase Probability Based on RFID Data in a Japanese Supermarket Reviewed
Keiji Takai, Katsutoshi Yada
KNOWLEDGE-BASED AND INTELLIGENT INFORMATION AND ENGINEERING SYSTEMS, PT III 6278 254 - 263 2010
-
Conditions for elimination of factor indeterminacy in time series factor analysis Reviewed
高井 啓二
New Trends in Psychometrics 443-452 2009
-
Simple Computation of Maximum Likelihood Estimates in Latent Class Model with Equality and Constant Constraints Reviewed
Keiji Takai, Yutaka Kano
COMMUNICATIONS IN STATISTICS-SIMULATION AND COMPUTATION 38 ( 3 ) 654 - 665 2009
-
Test of independence in a 2 x 2 contingency table with nonignorable nonresponse via constrained EM algorithm Reviewed
Keiji Takai, Yutaka Kano
COMPUTATIONAL STATISTICS & DATA ANALYSIS 52 ( 12 ) 5229 - 5241 2008.8
Books
-
欠測データの統計科学--医学と社会科学への応用--
高井啓二, 星野崇宏, 野間久史（ Role： Joint author）
岩波書店 2016.4
Research Projects
-
A Study of Cognitive Processes in Investor Decision Making Using Eye Tracking
Grant number：22H00888 2022.4 - 2026.3
Japan Society for the Promotion of Science Grants-in-Aid for Scientific Research Grant-in-Aid for Scientific Research (B)
Grant amount：\16640000 （ Direct Cost: \12800000 、 Indirect Cost：\3840000 ）
-
EMアルゴリズムに代わる欠測データを用いたパラメータ推定法の開発
Grant number：18K11205 2018.4 - 2023.3
日本学術振興会 科学研究費助成事業 基盤研究(C)
高井 啓二
Grant amount：\4160000 （ Direct Cost: \3200000 、 Indirect Cost：\960000 ）
本年度はいろいろな欠測データメカニズム下での情報量規準の開発と，完全にランダムな欠測の下での行列の一致推定の問題に取り組んだ．これまで開発されてきた情報量規準には応用上重大な問題点があった．それは情報量規準によって変数を選択するとき，欠測データメカニズムを一切考慮していないことであった．本来，欠測データメカニズムがランダムな欠測でない限り，情報量規準は適用できないはずである．しかし，従来の情報量規準を用いると，適用できないはずの欠測データメカニズムも選択してしまうことがあった．そこで，本年度は， 手に入る全てのデータを用いるときにはランダムな欠測であるという仮定をおいて，情報量規準を開発した．ここでおいた仮定は，変数が多ければ多いほど欠測を生じさせる変数がデータに含まれるという現実によく対応している．この仮定の下であれば，どのようなパラメータであっても一致推定可能になる．さらに，推定量の分散を小さくすることも可能となる．この情報量規準によって任意の変数のセットが選ばれても，パラメータは一致推定できるので，正確な情報量規準が計算できる．もう一つの研究では，完全にランダムな欠測の状況下での行列が，特にサンプルサイズが小さい場合に，非正定値になってしまう場合に対する対策法を開発した．このような問題が起きると，分散共分散行列が正定値であることを仮定している一般的な多変量解析法は一切使えなくなってしまう．対策としては，情報量をできるだけ減らさずに行列を正定値に変形することが重要になる．本年度は，非正定行列から最も近い正定値を固有値分解を用いて作成する方法を組み込んだアルゴリズムを作成した．
-
パスデータの融合による研究フロンティアの創出
Grant number：16H02034 2016.4 - 2021.3
日本学術振興会 科学研究費助成事業 基盤研究(A)
矢田 勝俊, 高井 啓二, 宮崎 慧, 石橋 健, 李 振, 里村 卓也, 金子 雄太, 中原 孝信, 左 毅, 市川 昊平
Grant amount：\38090000 （ Direct Cost: \29300000 、 Indirect Cost：\8790000 ）
多様なパスデータを用いた解析、モデル化を行うために、（1）消費者行動モデルの構築、（2）国際ワークショップの開催、（3）パスデータ融合の体系化に取り組み2019年度の計画通り研究を遂行することができた。
（1）2018年度に実施した店舗実験で収集したパスデータをもとに、様々な消費者行動モデルを構築した。セルフコントロールが店内購買行為に与える時系列の影響についてモデル化に成功、また視線追跡を統合し、個人の異質性を前提とした確率モデルを提示することができた。
（2）国際ワークショップの開催では、KES2019に併設する形で招待セッションを主催し、5本の発表を受入ながら、ブダペスト（ハンガリー）で行われた。顧客動線、視線追跡などの発表も含め、重要な意見の交換が行われた。そしてIEEEが主催する国際会議ICDMに併設する形で、Data mining for serviceという国際ワークショップを主催し、多くの発表を受入ながら、本提案の発表を行った。また、天野教授（ハーバード大学）やWedel教授（メリーランド大学）と議論しながら、体系化について貴重なアイデアを得ることができた。特にバースト検知手法を用いた混雑状況予測モデルについて、体系化が進み、論文化のめどをつけることができた。
（3）体系化については、上述したような海外の研究者との議論を通して体系化のアイデアをまとめつつ、本提案チームの中で議論を深めることができた。ただし、3月に実施予定であった今期、最後の打合せが新型コロナウイルス感染症の感染拡大をうけて中止になった。 -
How serious is nonignorable missingness?
Grant number：16K12402 2016.4 - 2018.3
Japan Society for the Promotion of Science Grants-in-Aid for Scientific Research Grant-in-Aid for Challenging Exploratory Research
Kano Yutaka, TAKAI Keiji, OTSU Tatsuo, MORIKAWA Kosuke, IMADA Miyuki, TAKAGI Yoshiharu, NAGASE Mario, Kim Jae-Kwang, Yuan Ke-Hai, Jamshidian Mortaza(Mori)
Grant amount：\3380000 （ Direct Cost: \2600000 、 Indirect Cost：\780000 ）
Under NMAR missingness, the observed likelihood, without a missing-data mechanism, leads to a biased MLE. In this research, we developed a new methodology to express the bias of the MLE due to the missingness in closed form. Using the formula, we provided several mathematical conditions under which inclusion of auxiliary variables reduces or inflates the bias.
The formula described above holds for any missing-data mechanism. This strong consequence can be proved because a shared-parameter model is taken for missingness.
A final contribution of this research to be reported is to take a semi-parametric way to relax the strong condition required for the conventional missing-data analysis. -
Study on the creation of a new ignobility condition and the investigation of an estimator distribution in the analysis of missing data
Grant number：26730022 2014.4 - 2018.3
Japan Society for the Promotion of Science Grants-in-Aid for Scientific Research Grant-in-Aid for Young Scientists (B)
TAKAI Keiji
Grant amount：\3770000 （ Direct Cost: \2900000 、 Indirect Cost：\870000 ）
In this project, I developed the theory for analysis of missing data and applied it to special data in the discriminant analysis. As theoretical study research, I derived conditional independences equivalent to MAR(missing at random) under monotonic and non-monotonic missing-data mechanisms. In addition, I constructed a method to overcome some difficulties in computation and estimation of the parameters of interest with missing data by using the selection matrix. It allows us to investigate properties of the estimator which are necessary for inference. As application research, I tackled a semi-supervised learning problem in discriminant analysis using the missing-data analysis theory. The semi-supervised learning is an estimation of the parameters from the partially observed data. I showed that the use of the missing-data analysis theory makes it possible to obtain the correct discriminant rule even with the such partially observed data.
-
New developments of missing data analysis: NMARness and APB
Grant number：25540011 2013.4 - 2016.3
Japan Society for the Promotion of Science Grants-in-Aid for Scientific Research Grant-in-Aid for Challenging Exploratory Research
Yutaka Kano, IWASAKI Manabu, TAKAI Keiji, OTSU Tatsuo, HIROSE Kei, KAMATANI Kengo, KIKUCHI Kenichi, Sobel Michael E., Yuan Ke-Hai, Ricardo Silva, Mortaza Jamshidian, Aapo Hyvarinen
Grant amount：\3640000 （ Direct Cost: \2800000 、 Indirect Cost：\840000 ）
This research project has been completed by the two research groups conducted by Professor Yutaka Kano and Professor Manabu Iwasaki. We have offered research colloquiums several times for each year to advance the research project. The aim of the research project is to re-structure the theory of missing data analysis and to apply them to some statistical models for the analysis with missing data. Results of the project include mathematically weakening the MAR condition, defining NMARness and Approximate population Bias (APB) and studying mathematical properties of the NMARness and APB. Applying these theoretical results, we studied effectiveness of introducing auxiliary variables in several statistical models for the analysis of missing data. One particular result is to derive mathematical conditions under which introducing surrogate endpoints can reduce the bias of the MLE for data with possibly missing data at the endpoint.
-
Development of Analysis Methods based on Generalized Methods of Moments for Nonignorable Missing Data
Grant number：24700284 2012.4 - 2014.3
Japan Society for the Promotion of Science Grants-in-Aid for Scientific Research Grant-in-Aid for Young Scientists (B)
TAKAI Keiji
Grant amount：\2860000 （ Direct Cost: \2200000 、 Indirect Cost：\660000 ）
As the first result of my study, I found that the maximum likelihood estimator (MLE) constructed from incomplete data has desirable properties such as consistency and asymptotic normality under the different conditions to the case in which completely observed data are available. The second result is application of the first result to discriminant analysis with partially labeled data. Since the partially labeled data can be regarded as missing data, the first result can also be applied to estimation of the parameters in such discriminant model when constructing a discriminant rule. It is found that all data available to us should be used when the observations used to construct the rule is completely randomly chosen, while there are times when not all observations with or without labels should be used for the data which are chosen depending on the value of the feature vector.
-
Experiment and Modeling of Consumer Behavior Analysis by Using Shopping Path Data
Grant number：22243033 2010.4 - 2015.3
Japan Society for the Promotion of Science Grants-in-Aid for Scientific Research Grant-in-Aid for Scientific Research (A)
YADA Katsutoshi, KAWAKAMI Tomoko, NAKAHARA Takanobu, ICHIKAWA Kouhei, NISHIOKA Kenichi, TAKAI Keiji
Grant amount：\32500000 （ Direct Cost: \25000000 、 Indirect Cost：\7500000 ）
The purpose of this research is to provide a model of consumer behavior which presents the relationship between consumer's in-store movement and purchasing behavior. We integrate customer's purchase history data with shopping path data which is consumer's in-store movement data in a retailer and then present various theoretical models such as customer's behavior model based on a string analysis, a time-series model of customer's movement and so on. Actually we implemented some marketing promotions in the real retailer based on our findings and achieved good sales performance. In conclusion we present a comprehensive framework of shopping path research based on these findings. We show significant implications for retail business and some theoretical contributions in marketing.
-
Statistical prediction, causation, incomplete data analysis and foundation of sciencee
Grant number：22300096 2010.4 - 2014.3
Japan Society for the Promotion of Science Grants-in-Aid for Scientific Research Grant-in-Aid for Scientific Research (B)
KANO YUTAKA, DEGUCHI Yasuo, WASHIO Takashi, HAMAZAKI Toshimitsu, TAKAGI Yoshiji, SUGIMOTO Tomoyuki, TAKAI Keiji, NAITO Kanta, SHIMIZU Shohei, KATAYAMA Shota, YAMAMOTO Michio, SONG Xinyuan, JAMSHIDIAN Mortaza, HYVARINEN Aapo, YUAN Ke-hai
Grant amount：\17810000 （ Direct Cost: \13700000 、 Indirect Cost：\4110000 ）
Analysis of incomplete data has been troublesome both theoretically and practically. In particular nonignorable missingness has been a serious issue in statistics. An alternative perspective of the theory of missing data analysis is to provide an insightful view of statistical causal inference. Some notable research outcomes include development of the analysis of doubly censored data, a new method of exploring causal structure for data with latent confounders via the LiNGAM approach, incomplete data analysis with a shared-parameter model and development of the EM algorithm with constraints.