Decision rule mining using classification consistency rate
Introduction
Rule induction is one of the most important techniques of machine learning, expert system, knowledge discovery and data mining. To handle this issue, many inductive learning methods, such as induction of decision trees [1], [2], rule induction methods [3], [4], [5], [6], [7] and rough set theory [8], [9], [10], [11], [12] are introduced and applied to extract knowledge from databases.
Rough set theory, introduced by Pawlak, is a useful mathematic approach for dealing with vague and uncertain information. It has attracted the attention of many researchers who have studied its theories and its applications during the last decades [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [25], [26], [27], [28], [29], [30], [31], [32], [33].
A number of approaches of rule induction based on rough set theory have been proposed. These approaches can be broadly divided into two categories. One is exhaustive category. Skowron [34], [35] proposed famous rule induction approaches based on discernibility matrix and boolean reasoning. For particular clinic application, Tsumoto [27], [28] introduced a rule induction approach called PRIMEROSE, which extracts not only classification rules but also other medical knowledge needed for diagnosis. The induction algorithm consists of two procedures, one is an exhaustive search procedure to induce the exclusive rule through all the attribute–value pairs, and the other is a postprocessing procedure to induce inclusive rules through the combination of all the attribute–value pairs. The other category is heuristic. Grzymala-Busse proposed famous LEM2 algorithm in LERS [9], [36], [37], [38], [39] which is a representative approach for rule induction by rough set theory. LEM2 explores the search space of the whole attribute–value pairs (called descriptors), and a pair who covers the most number of objects is selected to induce decision rule each time. In other words, it implements the rule induction procedure from the viewpoint of descriptors.
However, in some cases, rule induction from the viewpoint of attributes is more reasonable. For example, users may be interested in the rules about some appointed attribute or attribute sets. Based on this observation, we suggest the viewpoint of attribute space rather than viewpoint of descriptor space for the issue of rule induction in this paper. This viewpoint supplies us an approach to study the data at the understandable semantic level of attributes. From the viewpoint of attribute space, we propose a new strategy for decision rules induction from decision systems based on the concept of classification consistency rate which is defined in this paper to search the attribute space. The proposed rule mining strategy is denoted as DRICA (Decision Rule mIning using Consistency rAte).
The remainder of the paper is organized as follows. Some preliminaries about rough set theory are reviewed in Section 2. In Section 3, the algorithm of rule induction based on classification consistency is introduced. An example is used to illustrate the process of generating rules in Section 4. Experiments on several data sets are conducted to test the proposed approach, and the results are compared in Section 5. Section 6 concludes the paper.
Section snippets
Rough set theory
In this section, we first review some basic notions of rough set theory, which can also be referred to [8].
Classification consistency rate
Given a decision table DT = (U, C ∪ D, V, f), let us consider the formula
It represents the number of objects can be classified by attribute set P ⊆ C. Definition 2 Given a decision table DT = (U, C ∪ D, V, f), classification consistency rate relative to attribute set P ⊆ C can be defined as:
Classification consistency rate CCRP,D describes the classification ability of attribute set P ⊆ C relative to decision attribute D. Example 1 Table 1 contains four objects U = {x1, x2, x3, x4}, three condition attributes C = {a, b, c},
An illustrative example
In this section, an example is given to show how the proposed algorithm can be used to generate rules from decision tables. The same example will be also computed by LEM2 step by step to show the difference between the two approaches. Assume the data set is shown in Table 2.
Table 2 is a decision system with U = {1, 2, … , 9}, C = {a, b, c}, D = {d}. Since object 6 and 8 have the same condition attribute values but different decision attribute value, this decision system is inconsistent. The inconsistent
Experiment study
In last section, we describe the procedure of proposed algorithm to induce rules from inconsistent data set according to an example. In this section, empirical experiments are conducted to test the proposed algorithm.
Conclusion
In this paper, we propose a rough set approach for rule induction based on classification consistency rate called DRICA. DRICA implements the decision rule induction procedure from the viewpoint of attribute rather than descriptors. In each iteration step, it searches in the attribute space instead of descriptor space. It may obtain more than one rules at a single step. Since DRICA uses classification consistency rate as the criterion of selecting attributes, the rules with confidence of 1 rank
Acknowledgements
This work was supported by the National Natural Science Foundation of China (Nos. 61070074 and 60703038).
References (41)
PRISM: an algorithm for inducing modular rules
International Journal of Man-Machine Studies
(1987)A theory and methodology of inductive learning
Artificial Intelligence
(1983)A theory and methodology of inductive learning
- et al.
A soft-computing based rough sets classifier for classifying IPO returns in the financial markets
Applied Soft Computing
(2012) Rough 3-valued algebras
Information Sciences
(2008)- et al.
Uncertainty measurement for interval-valued decision systems based on extended conditional entropy
Knowledge-Based Systems
(2012) - et al.
Approximations and uncertainty measures in incomplete information systems
Information Sciences
(2012) - et al.
Attribute selection based on information gain ratio in fuzzy rough set theory with application to tumor classification
Applied Soft Computing
(2013) - et al.
Attribute selection based on a new conditional entropy for incomplete decision systems
Knowledge-Based Systems
(2013) - et al.
Enhancing evolutionary instance selection algorithms by means of fuzzy rough set based feature selection
Information Sciences
(2012)
A rough set approach to multiple dataset analysis
Applied Soft Computing
A new weighted rough set framework based classification for egyptian neonatal jaundice
Applied Soft Computing
Rough-set-based timing characteristic analyses of distance protective relay
Applied Soft Computing
Extraction of experts decision rules from clinical databases using rough set model
Journal of Intelligent Data Analysis
Approaches to attribute reductions based on rough set and matrix computation in inconsistent ordered information systems
Knowledge-Based Systems
Transformation of bipolar fuzzy rough set models
Knowledge-Based Systems
A comparative study of rough sets for hybrid data
Information Sciences
A general frame for intuitionistic fuzzy rough sets
Information Sciences
Reduction and axiomization of covering generalized rough sets
Information Sciences
Classification and Regression Trees
Cited by (40)
Information-theoretic measures of uncertainty for interval-set decision tables
2021, Information SciencesUncertainty measures for interval set information tables based on interval δ-similarity relation
2019, Information SciencesUncertainty learning of rough set-based prediction under a holistic framework
2018, Information SciencesUncertainty measurement for incomplete interval-valued information systems based on α-weak similarity
2017, Knowledge-Based SystemsCatoptrical rough set model on two universes using granule-based definition and its variable precision extensions
2017, Information SciencesCitation Excerpt :Rough set theory, proposed by Pawlak [34,35] has been conceived as an excellent tool to analyze and handle intelligent systems characterized by imprecise, vague and uncertain information in many fields, such as data mining, knowledge discovery, decision making and so on [5,6,8,9,12,19,21,24,26,37,38,43,44,49,50,54,57].
Rough sets in distributed decision information systems
2016, Knowledge-Based Systems