Hidenao Abe

Evaluation of rule interestingness measures in medical knowledge discovery in databases

OBJECTIVE We discuss the usefulness of rule interestingness measures for medical KDD through experiments using clinical datasets, and, based on the outcomes of these experiments, also consider how to utilize these measures in postprocessing. METHODS AND MATERIALS We first conducted an experiment to compare the evaluation results derived from a total of 40 various interestingness measures with those supplied by a medical expert for rules discovered in a clinical dataset on meningitis. We calculated and compared the performance of each interestingness measure to estimate a medical experts interest using f-measure and correlation coefficient. We then conducted a similar experiment for hepatitis. RESULTS AND CONCLUSION The comprehensive results of experiments on meningitis and hepatitis indicate that the interestingness measures, accuracy, chi-square measure for one quadrant, relative risk, uncovered negative, and peculiarity, have a stable, reasonable performance in estimating real human interest in the medical domain. The results also indicate that the performance of interestingness measures is influenced by the certainty of a hypothesis made by the medical expert, and that the combinational use of interestingness measures will contribute to support medical experts to generate and confirm their hypotheses through human-system interaction.

Implementing an integrated time-series data mining environment based on temporal pattern extraction methods: a case study of an interferon therapy risk mining for chronic hepatitis

In this paper, we present the implementation of an integrated time-series data mining environment. Time-series data mining is one of key issues to get useful knowledge from databases. With mined time-series patterns, users can aware not only positive results but also negative result called risk after their observation period. However, users often face difficulties during time-series data mining process for data pre-processing method selection/construction, mining algorithm selection, and post-processing to refine the data mining process as other data mining processes. It is needed to develop a time-series data mining environment based on systematic analysis of the process. To get more valuable rules for domain experts from a time-series data mining process, we have designed an environment which integrates time-series pattern extraction methods, rule induction methods and rule evaluation methods with active human-system interaction. After implementing this environment, we have done a case study to mine time-series rules from blood and urine biochemical test database on chronic hepatitis patients. Then a physician has evaluated and refined his hypothesis on this environment. We discuss the availability of how much support to mine interesting knowledge for an expert.

Trend detection from large text data

In temporal text mining, some importance indices such as simple appearance frequency, tf-idf, and differences of some indices play the key role to point out remarkable trends of terms in sets of documents. However, almost of conventional methods have treated their remarkable trends as discrete statuses for each time-point or fixed period. In this paper, we present a method to find out remarkable temporal behaviors of technical terms by using several importance indices and temporal clustering on the indices. The implemented method with three indices and k-means clustering performed on research document sets. The results of the case study show that the method has a feasibility to point out emergent, popular, and subsiding terms based on the linear trend of the temporal clusters of the technical terms.

A rule evaluation support method with learning models based on objective rule evaluation indexes

In this paper, we present a novel rule evaluation support method for post-processing of mined results with rule evaluation models based on objective indexes. Post-processing of mined results is one of the key issues to make a data mining process successfully. However, it is difficult for human experts to evaluate many thousands of rules from a large dataset with noises completely. To reduce the costs of rule evaluation procedures, we have developed the rule evaluation support method with rule evaluation models, which are obtained with objective rule evaluation indexes and evaluations of a human expert for each rule. Since the method is needed more accurate rule evaluation models, we have compared learning algorithms to construct rule evaluation models with the actual meningitis data mining result and actual rule sets from UCI datasets. Then we show the availability of our adaptive rule evaluation support method.

Developing an Integrated Time-Series Data Mining Environment for Medical Data Mining

In this paper, we present an integrated time-series data mining environment for medical data mining. Medical time-series data mining is one of key issues to get useful clinical knowledge from medical databases. However, users often face difficulties during such medical time-series data mining process for data preprocessing method selection/construction, mining algorithm selection, and post-processing to refine the data mining process as shown in other data mining processes. To get more valuable rules for medical experts from a time-series data mining process, we have designed an environment which integrates time- series pattern extraction methods, rule induction methods and rule evaluation methods with visual human-system interface. After implementing this environment, we have done a case study to mine time- series rules from blood/urine biochemical test database on chronic hepatitis patients. The result shows the availability to find out valuable clinical course rules based on time-series pattern extraction. Furthermore, we compared the difference of time-series pattern extraction methods with objective rule evaluation results.

Numerical time-series pattern extraction based on irregular piecewise aggregate approximation and gradient specification

This paper proposes and evaluates a method for extracting interesting patterns from numerical time-series data which takes account of user subjectivity. The proposed method conducts irregular sampling on the data preserving the subjectively noteworthy features using a user specified gradient. It also conducts irregular quantization, preserving the intrinsically objective characteristics of the data using statistical distributions. It then extracts representative patterns from the discretized data using group average clustering. Experimental results using benchmark datasets indicate that the proposed method does not destroy the intrinsically objective features, since it has the same performance as the basic subsequence clustering using K-Means algorithm. Results using a dataset from a clinical hepatitis study indicate that it extracts interesting patterns for a medical expert.

Text Categorization with Considering Temporal Patterns of Term Usages

In document categorization method by using similarity measures based on word vectors, it is important to determine key words to characterize each document. However, conventional methods select the key words based on their frequency or/and particular importance index such as tf-idf. In this paper, we propose a method to characterize each document by using temporal clusters of technical term usages. The method obtains document clusters based on the similarity between the document that are characterized by the temporal patterns of an importance index for considering temporal differences of the term usages In the experiment, we compare document categorization results based on document clustering by using the two types of feature sets about two sets of bibliographical documents. By regarding to the experimental results, we discuss the usefulness of the temporal patterns of term usages to characterize the documents.

Analyzing Behavior of Objective Rule Evaluation Indices Based on a Correlation Coefficient

In this paper, we present an analysis of behavior of objective rule evaluation indices on classification rule sets using Pearson productmoment correlation coefficients. To support data mining post-processing, which is one of important procedures in a data mining process, at least 40 indices are proposed to find out valuable knowledge. However, their behavior have never been clearly articulated. Therefore, we carried out a correlation analysis between each objective rule evaluation index. In this analysis, we calculated average values of each index using bootstrap method on 32 classification rule sets learned with information gain ratio. Then, we found the following relationships based on the correlation coefficient values: similar pairs, discrepant pairs, and independent indices. With regarding to this result, we discuss about relative functional relationships between each group of objective indices.

Comparing similarity of concepts identified by temporal patterns of terms in biomedical research documents

In this paper, we present an analysis of a relationship between temporal trends of automatically extracted terms in medical research document and their similarities on a structured vocabulary. In order to obtain the temporal trends, we used our temporal pattern extraction method that combines an automatic term extraction, an importance index of the terms, and clustering for the values in each period. By using a set of medical research documents that were published every year, we extracted temporal patterns of the automatically extracted terms. Then, we calculated their similarities on the medical taxonomy by defining a distance on the tree structure. For analyzing the relationship between the terms included in the patterns and the similarity of the terms on the taxonomy, the differences of the averaged similarities of the terms in each pattern are compared between the two trends of the temporal patterns.

Detecting Temporal Trends of Technical Phrases by Using Importance Indices and Linear Regression

In this paper, we propose a method for detecting temporal trends of technical terms based on importance indices and linear regression methods. In text mining, importance indices of terms such as simple frequency, document frequency including the terms, and tf-idf of the terms, play a key role for finding valuable patterns in documents. As for the documents, they are often published daily, monthly, annually, and irregularly for each purpose. Although the purposes of each set of documents are not changed, roles of terms and the relationship among them in the documents change temporally. In order to detect such temporal changes, we combined a method to extract terms, importance indices of terms, and trend identification based on linear regression analysis. Empirical results show that our method detected emergent and subsiding trends of extracted terms in a corpus of a research domain. By comparing this method with the existing burst detection method, we investigated the trend of phrases consisting of several burst words in the titles of AAAI and IJCAI.