Wyniki wyszukiwania - Biblioteka Nauki

1

Contingency Matrix Theory II: Degree of Dependence as Granularity

100%

Tsumoto S. , Hirano S.

Fundamenta Informaticae

|

2009

|

tom Vol. 90, nr 4

427-442

EN

The degree of granularity of a contingency table is closely related with that of dependence of contingency tables. We investigate these relations from the viewpoints of determinantal devisors and determinants. From the results of determinantal divisors, it seems that the devisors provide information on the degree of dependencies between the matrix of the whole elements and its submatrices and the increase of the degree of granularity may lead to that of dependence. However, the other approach shows that a constraint on the sample size of a contingency table is very strong, which leads to the evaluation formula where the increase of degree of granularity gives the decrease of dependency

2

Statistical Independence and Determinants in a Contingency Table - Interpretation of Pearson Residuals based on Linear Algebra -

100%

Tsumoto S. , Hirano S.

Fundamenta Informaticae

|

2009

|

tom Vol. 90, nr 3

251-267

EN

This paper analyzes pearson residuals, which is an important element of chi-square test statistic, in a contingency table from the viewpoint of matrix theory as follows. First, a given contingency table is viewed as a matrix and the residual of each element in a matrix are obtained as the difference bewteen observed values and expected values calculated by marginal distributions. Then, each residual σi,j/sub> is decomposed into the linear sum of the 2× 2 subderminants of a original matrix, except for i-th column and j-th row. Furthermore, the number of the determinants is equal to the degree of freedom for the chi-square test statistic for a given contingency table. Thus, 2 × 2 subdeterminants in a contingencymatrix determine the degree of statistical independence of two attributes as elementary granules.

3

Visualization of Differences between Rules' Syntactic and Semantic Similarities using Multidimensional Scaling

100%

Tsumoto S. , Hirano S.

Fundamenta Informaticae

|

2007

|

tom Vol. 78, nr 4

561-573

EN

One of the most important problems with rule induction methods is that it is very difficult for domain experts to check millions of rules generated from large datasets, although the discovery from these rules requires deep interpretation from domain knowledge. Although several solutions have been proposed in the studies on data mining and knowledge discovery, these studies are not focused on similarities between rules obtained. When one rule r1 has reasonable features and the other rule r2 with high similarity to r1 includes unexpected factors, the relations between these rules will become a trigger to the discovery of knowledge. In this paper, we propose a visualization approach to show the similarity relations between rules based on multidimensional scaling, which assign a two-dimensional cartesian coordinate to each data point from the information about similarities between this data and others data. We evaluated this method on two medical data sets, whose experimental results show that knowledge useful for domain experts can be found.

4

Incremental Induction of Medical Diagnostic Rules Based on Incremental Sampling Scheme and SubRule Layers

100%

Tsumoto S. , Hirano S.

Fundamenta Informaticae

|

2013

|

tom Vol. 127, nr 1-4

209--223

EN

This paper proposes a new framework for incremental rule induction of medical diagnostic rules based on incremental sampling scheme and rule layers. When an example is appended, four possibilities can be considered. Thus, updates of accuracy and coverage are classified into four cases, which give two important inequalities of accuracy and coverage for induction of probabilistic rules. By using these two inequalities, the proposed method classifies a set of formulae into four layers: the rule layer, subrule layer (in and out) and the non-rule layer. Then, the obtained rule and subrule layers play a central role in updating proabilistic rules. If a new example contributes to an increase in the accuracy and coverage of a formula in the subrule layer, the formula is moved into the rule layer. If this contributes to a decrease of a formula in the rule layer, the formula is moved into the subrule layer. The proposed method was evaluated on a dataset regarding headaches, whose results show that the proposed method outperforms the conventional methods.

5

Risk Mining in Medicine: Application of Data Mining to Medical Risk Management

100%

Tsumoto S. , Hirano S.

Fundamenta Informaticae

|

2010

|

tom Vol. 98, nr 1

107-121

EN

This paper proposes an application of data mining to medical risk management, where data mining techniques are applied to detection, analysis and evaluation of risks potentially existing in clinical environments. We applied this technique to the following two medical domains: risk aversion of nurse incidents and infection control. The results show that data mining methods were effective to detection and aversion of risk factors.

6

Maintenance and Discovery of Domain Knowledge for Nursing Care using Data in Hospital Information System

80%

Iwata H. , Hirano S. , Tsumoto S.

Fundamenta Informaticae

|

2015

|

tom Vol. 137, nr 2

237--252

EN

[Introduction] Schedule management of hospitalization is important to maintain or improve the quality of medical care and application of a clinical pathway is one of the important solutions for the management. Although several kinds of deductive methods for construction for a clinical pathway have been proposed, the customization is one of the important problems. This research proposed an inductive approach to support the customization of existing clinical pathways by using data on nursing actions stored in a hospital information system. [Method] The number of each nursing action applied to a given disease during the hospitalization was counted for each day as a temporal sequence. Temporal sequences were compared by using clustering and multidimensional scaling method in order to visualize the similarities between temporal patterns of clinical actions. [Results] Clustering and multidimensional scaling analysis classified these orders to one group necessary for the treatment for this DPC and the other specific to the status of a patient. The method was evaluated on data sets of ten frequent diseases extracted from hospital information system in Shimane University Hospital. Cataracta and Glaucoma were selected. Removing routine and poorly documented nursing actions, 46 items were selected for analysis. [Discussion] Counting data on executed nursing orders were analyzed as temporal sequences by using similarity-based analysis methods. The analysis classified the nursing actions into the two major groups: one consisted of orders necessary for the treatment and the other consisted of orders dependent on the status of admitted patients, including complicated diseases, such as DM or heart diseases. The method enabled us to inductive construction of standardized schedule management and detection of the conditions of patients difficult to apply the existing or induced clinical pathway.