Ahmed Samet

Mining Frequent Itemsets in Evidential Database

Mining frequent patterns is widely used to discover knowledge from a database. It was originally applied on Market Basket Analysis (MBA) problem which represents the Boolean databases. In those databases, only the existence of an article (item) in a transaction is defined. However, in real-world application, the gathered information generally suffer from imperfections. In fact, a piece of information may contain two types of imperfection: imprecision and uncertainty. Recently, a new database representing and integrating those two types of imperfection were introduced: Evidential Database. Only few works have tackled those databases from a data mining point of view. In this work, we aim to discuss evidential itemset’s support. We improve the complexity of state of art methods for support’s estimation. We also introduce a new support measure gathering fastness and precision. The proposed methods are tested on several constructed evidential databases showing performance improvement.

Classification with Evidential Associative Rules

Mining database provides valuable information such as frequent patterns and especially associative rules. The associative rules have various applications and assets mainly data classification. The appearance of new and complex data support such as evidential databases has led to redefine new methods to extract pertinent rules. In this paper, we intend to propose a new approach for pertinent rule’s extraction on the basis of confidence measure redefinition. The confidence measure is based on conditional probability basis and sustains previous works. We also propose a classification approach that combines evidential associative rules within information fusion system. The proposed methods are thoroughly experimented on several constructed evidential databases and showed performance improvement.

Evidential data mining: precise support and confidence

Associative classification has been shown to provide interesting results whenever of use to classify data. With the increasing complexity of new databases, retrieving valuable information and classifying incoming data is becoming a thriving and compelling issue. The evidential database is a new type of database that represents imprecision and uncertainty. In this respect, extracting pertinent information such as frequent patterns and association rules is of paramount importance task. In this work, we tackle the problem of pertinent information extraction from an evidential database. A new data mining approach, denoted EDMA, is introduced that extracts frequent patterns overcoming the limits of pioneering works of the literature. A new classifier based on evidential association rules is thus introduced. The obtained association rules, as well as their respective confidence values, are studied and weighted with respect to their relevance. The proposed methods are thoroughly experimented on several synthetic evidential databases and showed performance improvement.

Reliability estimation with Extrinsic and Intrinsic measure in belief function theory

Belief function theory provides a robust framework for uncertain information modeling. It also offers several fusion tools in order to profit from multi-source context. Nevertheless, fusion is a sensible task where conflictual information may appear especially when sources are unreliable. In belief function theory, a classical approach would estimate the sources reliability before any discounting operation. Existing solutions for sources reliability estimation, are based on the assumption that distance is the only factor for conflictual situations. Indeed, integrating only distance measures to estimate sources reliability is not sufficient where sources confusion may also be considered as conflict origin. In this paper, we tackle reliability estimation and we introduce a new discounting operator that considers those two possible conflict origins. The proposed approach is applied on benchmark data for classification purpose.

Evidential data mining for length of stay (LOS) prediction problem

Hospitals need to optimize their healthcare planning and organization to minimize costs. The indicator that is often used to measure the efficiency in hospital is the average length of stay. Many studies show a strong and obvious correlation between the costs of patients and the impatient Length Of Stay (LOS). In this paper, We propose to apply data mining techniques to predict the LOS. An evidential variant of data mining, called also evidential data mining, have been used to reduce the impact of uncertainty and missing data. New measures of itemset support and association rule confidence are applied. We introduce the Evidential Length Of Stay prediction Algorithm (ELOSA) that allow the prediction of the length of stay of a new patient. Therefore, the inpatient length of stay (LOS) can be predicted efficiently, the planning and management of hospital resources can be greatly enhanced. The proposal is evaluated on a real hospital dataset using 270 patient traces.

Classification of high-resolution remote sensing image by adapting the distance belief function estimation model

The multi-source information holds a great importance in processing complex and imprecise data. Unfortunately, it requires an adequate formalism capable to modelize and to fuse several information. The evidence theory distinguishes from all formalism by its capacity to modelize and treat imprecise and imperfect data. In this context, the high resolution images represent a huge amount of data and needs multi-source information to perform pattern recognition. In this paper, we present an adaption of the distance operator introduced by Denoeux for estimating belief functions. This proposed approach will be used to classify forest image remote sensing by identifying the tree crown classes.

Generic discounting evaluation approach for urban image classification

Belief function theory provides a robust framework for uncertain information modeling. It also offers several fusion tools in order to profit from multi-source context. Nevertheless, fusion is a sensible task where conflictual information may appear especially when sources are unreliable. Therefore, measuring sources reliability has been the center of many research and development. Existing solutions for sources reliability estimation are based on the assumption that distance is the only factor for conflictual situations. Indeed, integrating only distance measures to estimate sources reliability is not sufficient where sources confusion may be also considered as conflict origin. In this paper, we tackle reliability estimation and we introduce a new discounting operator that considers those two possible conflict origins. We propose an automatic method for discounting factor calculation. Those factors are integrated on belief classifier and tested on high-resolution image classification problem.

Reliability Estimation Measure: Generic Discounting Approach

In the belief function theory, several measures of uncertainty have been introduced. One of their possible use is unreliable source discounting before the fusion stage. Two different measures of uncertainty exist which are the intrinsic and extrinsic ones. The intrinsic measure makes it possible to assess the sources confusion whereas the extrinsic one measures the contradiction between sources. In this paper, we associate both measures in order to estimate the global reliability of a source. This method, named Generic Discounting Approach (GDA), is proposed in two different versions: Weighted GDA and Exponent GDA. Those reliability measures are integrated into a classifier. The method was tested, against to some pioneer approaches, on several UCI datasets as well as on an urban image classification problem and showed very encouraging results.

Integration of Extra-Information for Belief Function Theory Conflict Management Problem Through Generic Association Rules

Decision making by considering multiple information sources could provide interesting results. For that reason, fusion formalisms were a major concern in the belief function community. In this context, the Belief function theory allows information fusion thanks to its combinations tools that it integrates. Nevertheless, belief function theory highlights a limit in the merging of contradictory (conflictual) sources. Many authors tackled this problem offering contributions in this field. Unfortunately, no proposed operator has distinguished by its adequacy regardless the type of handled sources. In this paper, we demonstrate the limits of some referenced works and we diagnostic the issues origin. We propose a conflict management approach based on an extra-information that guides the treatment. We also integrate a generic associative base borrowed from the data mining domain in order to apply the adequate conflict management.

Belief Function Classification with Conflict Management: Application on Forest Image

Treating imprecise and uncertain data requires an adequate formalism allowing a fit modelization. Several formalisms can be identified such as Bayesian theory, fuzzy set theory and belief function theory. The belief function theory provides an adequate formalism to manipulate those imperfect data. It also allows source fusion thanks to the combination operators that it integrates. The fusion process generates an empty set mass denoted conflict that illustrates the contradiction rate between considered sources. In this work, we tackle the classification of a forest high-resolution remote-sensing image problem. In order to classify this image, we handled imperfect information with the belief function theory. We propose a method for classification based on belief function theory and source fusion. The introduced Redistributing Conflict Classification Approach (RCCA) analyzes the conflict resulting from the fusion and redistributes it to the most pertinent classes. An experimental comparison to well known literature classifiers is provided.