Abstracts of the papers presented at IDAMAP-97

Cungen Cao, Tze-Yun Leon, Adrian Pheng Kheong Leong, and Francis Seow Choen
Learning conditional probabilities for dynamic influence views

Dynamic decision making concerns problems in which both time and uncertainty are explicitly considered. A major challenge in applying decision analysis in dynamic decision problems is to elicit, estimate, and specify the numerous time-dependent conditional probabilities in the models. Based on DynaMoL (a Dynamic decision Modeling Language) framework, we examine the critical issues in automated learning of numerical parameters from large medical databases. In this paper, we present a Bayesian method for learning conditional probabilities from data for influence views, a key decision-modeling facility in DynaMoL, analyze how to elicit prior probabilities from the domain expert, and discuss several important issues on processing and preparing raw data for application in dynamic decision modeling.

Riccardo Bellazzi, Cristiana Larizza, and Albert Riva
Temporal abstractions for pre-processing and interpreting diabetes monitoring time abstractions

In this paper we describe a number of intelligent data analysis techniques to pre-process and analyze data coming from home monitoring of diabetic patients. In particular, we show how the combination of temporal abstractions with statistical and probabilistic techniques may be applied to derive useful summaries of patients' behaviour over a certain monitoring period. Finally, we describe how Intelligent Data Analysis methods may be used to index past cases to perform a case-based retrieval in a data-base of past cases.
Anders Hansson and Silvia Miksch
On-line identification of a patient-disease model for mechanical ventilation

Monitoring and therapy planning in real-world environments highly depend on good patient-disease models. The improvement of the technical equipment in modern intensive care units enables a huge number of on- and off-line data, which results in an information overload of the medical staff. Additionally, the underlying medical structure-function models are poorly understood or not applicable due to incomplete knowledge. We have developed an on-line identification scheme, which utilizes a priori knowledge as well as on-line measurements to identify the parameters of a disease model for mechanically ventilated newborn infants. The scheme benefits from an exponential weighting function to classify more recent measurement values as more important. We have evaluated our identification scheme with real medical data sets showing the benefits and drawbacks of our approach.
Du Junping, K.L. Rasmussen, J. Aagaard, Brian H. Mayoh, and Tom Sorensen
Applications of machine learning: A medical follow up study

This paper describes preliminary work that aims to apply some learning strategies to a medical follow-up study. An investigation of the application of three machine learning algorithms 1R, FOIL and InductH to identify the risk factors that govern the colposuspension cure rate has been made. The goal of this study is to induce generalised description or explanation of the classification attribute, colposuspension cure rate (completely cured, improved, unchanged and worse) from the 767 examples in the questionnaires. We looked for a set of rules that described which risks factors result in differences of cure rate. The results were encouraging, and indicate that machine learning can play a useful role in large scale medical problem solving.
Nada Lavrac, Elpida Keravnou, and Blaz Zupan
Intelligent data analysis in medicine

Excessive amounts of knowledge and data stored in medical databases request the development of specialized tools for storing and accessing of data, data analysis, and effective use of stored knowledge and data. This paper first sketches the history of research that led to the development of current intelligent data analysis techniques, aimed at narrowing the increasing gap between data gathering and data comprehension. Next, we present our view on the relation of Intelligent data analysis to Knowledge discovery in databases and Data mining. Finally, we discuss the need for intelligent data analysis in medicine and present the aims of research in this area.
Bing Liu, Wynne Hsu, and Shu Chen
Discovering conforming and unexpected classification rules

One problem in applying machine learning and knowledge discovery techniques to solve real-world problems is how to incorporate the user's concepts about the application domain into the learning process to discover interesting rules to the user. Rules are interesting if they are useful and/or provide new knowledge to the user. Interesting rules are subjective because they depend on the individual user's existing knowledge (concepts or hypotheses) about the domain and his/her interests at a particular point in time. In the applications of classification rule induction techniques to real-world problems, we encounter a number of problems regarding the production of interesting knowledge to the user. In this paper, we address a specific problem, i.e., characterizing the conforming and unexpected tuples in the database with respect to the user's existing concepts, which can be correct, partially correct or entirely incorrect. This helps the user to find interesting rules and enables him/her to have a better understanding of the domain.
Yuval Shahar
Knowledge-based interpolation of time-oriented clinical data

Temporal interpolation is the task of bridging gaps between time-oriented clinical data or abstracted concepts in a context-sensitive manner. It is one of the subtasks important for solving the temporal-abstraction task--abstraction of interval-based, higher-level concepts from time-stamped clinical data. We present a knowledge-based approach to the temporal-interpolation task. The temporal-interpolation mechanism we discuss relies, among other knowledge types, on a temporal-persistence model. This model employs local temporal-persistence functions that are temporally bidirectional (i.e., extend a belief measure in a predicate both into the future and into the past) and global, maximal-gap temporal-persistence functions that bridge gaps between interval-based predicates. We investigate the quantitative and qualitative properties implied by both types of persistence functions. Our goal is to formulate the knowledge required for solving the temporal-abstraction task, and in particular the temporal-interpolation subtask, so as to facilitate the acquisition of that knowledge, its maintenance, its reuse for the same task in different domains, and its sharing among different applications in the same domain. We have implemented our approach and evaluated it in several clinical domains.
Igor Zelic, Igor Kononenko, Nada Lavrac, and Vanja Vuga
Induction of decision trees and Bayesian classification applied to diagnosis of sport injuries

Machine learning techniques can be used to extract knowledge from data stored in medical databases. In our application, various machine learning algorithms were used to extract diagnostic knowledge to support the diagnosis of sport injuries. The applied methods include variants of the Assistant algorithm for top-down induction of decision trees, and variants of the Bayesian classifier. The available dataset was insufficent for reliable diagnosis of all sport injuries considered by the system. Consequently, expert-defined diagnostic rules were added and used as pre-classifiers or as generators of additional training instances for injuries with few training examples. Experimental results show that the classification accuracy and the explanation capability of the naive Bayesian classifier with the fuzzy discretization of numerical attributes was superior to other methods and was estimated as the most appropriate for practical use.
page maintained by Blaz Zupan, last update Aug 19, 1997