Random forests are powerful classifiers, but their complexity sometimes makes them difficult to interpret. In this paper, a method to extract and optimize decision rules from random forests using two key criteria: rule length (simplicity) and support (coverage), is proposed. Four optimization strategies based on length, support, and their combinations were employed in processing six benchmark datasets. The results show that pruning decision trees before rule extraction leads to shorter and more general rules. Among all strategies, optimizing for support followed by length consistently achieves the best accuracy while maintaining interpretability. This approach offers a practical way to create rule-based models that are both accurate and easier to understand.

The subject of the paper concerns the problem of deriving decision rules from distributed data. The paper examines issues of learning general and inner decision rules from a set of decision trees. Inner rules refer to the routes within decision trees from the root to leaf nodes , while general rules are arbitrary rules derived from attributes found in the set of decision trees. The paper illustrates that the optimization of general decision rules is NP-hard problem, so the authors propose heuristics H 1 and H 2 for this issue. Taking into account induction and optimization of inner decision rules an algorithm A is employed. Additionally, an approach based on global optimization relative to length, support, and sequential optimization is proposed. The presented algorithms were studied considering two perspectives (i) knowledge discovery from data and (ii) knowledge representation. In the first case, it is possible to discover patterns from the data and verify the induced model, in the second case, it is possible to represent knowledge in a comprehensible and explainable way. These elements are important in an era of heterogeneous, distributed data sources . Experiments were carried out on selected datasets from UCI ML and Kaggle repositories. In order to create a distributed data structure, an approach based on reducts induced by a genetic algorithm was employed. Obtained results show that there are cases where the global rule-based classifiers built in the framework of optimization of inner decision rules perform better in terms of accuracy than that of local models induced directly from subtables. In the case of algorithms H 1 and H 2 , the low complexity of models based on decision rules induced from a set of decision trees is noted.

Knowledge discovery is one of the key areas in predictive data mining tasks. Performing Classification tasks on a single source of data using a decision tree algorithm is a relatively straightforward process. However, the complication arises when we have distributed sources of data that yield sets of decision trees. Classifier ensembles are often created, and a decision is assigned to a new object based on a certain voting strategy. The article proposes a different approach to creating a rule-based classifier. Using a set of decision trees, a global model of decision rules is induced. It contains rules which are true for the maximum number of trees from a set of decision trees. This model is verified by data corresponding to distributed local data sources. The bootstrapping technique was used to obtain distributed data sources. Pruning of decision trees was applied to improve the accuracy of the rule-based classifier. The conducted experiments confirm the validity of using algorithm A for the learning of decision rules from a set of decision trees.

The research work aimed to analyse and compare five different feature selection methods belonging to filter and wrapper approaches. The methods were evaluated taking into account the knowledge discovery and knowledge representation perspectives. The Gradient Boosting classifier was used as a benchmark model due to its performance and flexibility. Extensive tests were conducted on the Wisconsin Breast Cancer Diagnostic dataset and indicated cases of enhanced prediction for the reduced set of features.

Estimation of importance for considered features is an important issue for any knowledge exploration process and it can be executed by a variety of approaches. In the research reported in this study, the primary aim was the development of a methodology for creating attribute rankings. Based on the properties of the greedy algorithm for inducing decision rules, a new application of this algorithm has been proposed. Instead of constructing a single ordering of features, attributes were weighted multiple times. The input datasets were discretised with several algorithms representing supervised and unsupervised discretisation approaches. Each resulting discrete data variant was exploited to construct a ranking of attributes. The effectiveness of the obtained rankings was confirmed through a rule filtering process governed by weighted attributes. The methodology was applied to the stylometric task of authorship attribution. The experimental outcomes demonstrate the value of the proposed research method, as it generally led to improved predictions while taking into account a noticeably decreased sets of attributes and decision rules.

The paper presents investigations concerning the decision rule filtering process controlled by the estimated relevance of available attributes.In the conducted study, two search directions were used, sequential forward selection and sequential backward elimination.The steps of sequential search were governed by three rankings obtained for variables, all related to characteristics of data and rules that can be induced, as follows, (i) a ranking based on the weighting factor referring to the occurrence of attributes in generated decision reducts, (ii) the OneR ranking exploiting short rule properties, and (iii) the proposed ranking defined through the operation of greedy algorithm for rule induction.The three rankings were confronted and compared from the perspective of their usefulness for the selection of rules performed in the two directions and with two strategies for rule selection.The resulting sets of rules were analysed with respect to the properties of the constituent decision rules and from the point of performance for all constructed rulebased classifiers.Substantial experiments were carried out in the stylometric domain, treating the task of authorship attribution as classification.The results obtained indicate that for all three rankings and search paths it was possible to obtain a noticeable reduction of attributes while at least maintaining the power of inducers, at the same time improving characteristics of rule sets.

The paper aims to analyze the expectations of employers towards the competencies possessed by graduates that arise along with the digital transformation processes. It presents the result of research in which the diagnostic survey method was used. In research, 144 University stakeholders from various Polish companies have participated. The study identified the expected competencies university graduates should have in the mentioned process and issues necessary for their development, which should form the basis of modern fields of study.

Decision rules belong to known forms of knowledge representation. Among popular measures of their quality length and support can be distinguished. Shorter rules are easier to understand and interpret. Support allows to present patterns hidden in the data. Nowadays, data mining tasks are oriented toward extracting knowledge from data in both distributed and centralized forms. Learning decision rules from a decision tree is a relatively simple task. However, the challenge arises when decision rules are induced from a set of decision trees. Moreover, in the case of distributed data, the decision trees may be constructed independently on different sources, and merging them into a unified set requires resolving conflicts and inconsistencies. In this paper, decision rules are constructed from distributed data based on decision trees induced using the randomly chosen attributes as the splitting criterion. The aim of the study is to compare the quality of two algorithms for constructing rules which are true for a maximum number of trees. The comparison was made based on three factors: the number of trees for which the rule is true, their length and support. Based on performed experiments it was possible to see that the number of true rules for the maximum number of decision trees from the set is greater for algorithm A than for heuristics H. This algorithm allows the induction of shorter rules with greater support compared to heuristic H. However, it should be also noted that the rules induced by heuristic H are often true for a larger number of trees than the rules constructed by algorithm A. Thus, both algorithms can be applied to distributed data.

Association rules are popular form for knowledge discovery domain. They are used for finding interesting relationships and patterns hidden in large data sets and in the area of associative classification, where usually rules with one item in the right-hand side are considered. There are many different approaches and algorithms for mining association rules. One of the most popular group are methods which are based on mining frequent itemsets, usually applied for data in transaction format. Such data can be transformed to binary information system which corresponds to matrix data format. Technological development means that we are dealing with an increasing amount of data that can be heterogeneous, taking into account their format and location. In this paper, we assume that dispersed data is represented by a finite set S of information systems with equal sets of attributes. We discuss one of the possible ways to the study association rules common to all information systems from the set S: building a joint information system for which the set of true association rules that are realizable for a given row r and have given attribute f on the right-hand side coincides with the set of association rules that are true for all information systems from S, are realizable for the row r, and have the attribute f on the right-hand side. We show how to build a joint information system in a polynomial time. When we build such an information system, we can apply to it various association rule learning algorithms.

The area of feature selection methods constantly expands along with the development of artificial intelligence domain, and has great impact on almost every field, whenever data is processed and explored. The paper presents research where a ranking method was proposed, inspired by an approach which comes from an algorithm for induction of decision rules. The ranking procedure was based on calculation of standard deviation for attributes, taking into account assigned class labels. This method was compared with another ranking mechanism, a modified version of popular Relief algorithm, with incorporating characteristics of variables by supervised discretisation. Comparison of obtained results included the aspect of knowledge representation as well as the perspective of the accuracy for constructed rule-based classifiers. The experiments were performed on datasets from stylometry domain, where authorship attribution was considered as a classification task, and stylometric descriptors as characteristic features defining writing styles of authors.

Decision and association rules are known and popular forms of knowledge representation. They are used in many areas of data mining and have different applications. One of them is associative classification which finds association rules that have only class label in the consequent part and satisfies the minimum support and confidence thresholds. The aim of this paper is to merge set of association rules induced by Apriori algorithm with the set of decision rules induced by selected heuristics, forming a combined classifier. Experimental results concern parameters of induced sets of rules as number of unique rules, their length and support, and classification accuracy involving two voting strategies as decision list and standard voting. It was shown, that for considered parameters of Apriori algorithm, the combined classifier allows to obtain improved classification accuracy compared to the classifier based only on association rules. Therefore selected heuristic for induction of decision rules can be applied in associative classification tasks.

When mining of input data is focused on rule induction, knowledge, discovered in exploration of existing patterns, is stored in combinations of certain conditions on attributes included in rule premises, leading to specific decisions. Through their properties, such as lengths, supports, cardinalities of rule sets, inferred rules characterise relations detected among variables. The paper presents research dedicated to analysis of these dependencies, considered in the context of various discretisation methods applied to the input data from stylometric domain. For induction of decision rules from data, Classical Rough Set Approach was employed. Next, based on rule properties, several factors were proposed and evaluated, reflecting characteristics of available condition attributes. They allowed to observe how variables and rule sets changed depending on applied discretisation algorithms.

Decision rules are popular form of knowledge representation. From this point of view, length of such rules is an important factor since it influences on data understanding by experts. Unfortunately, the problem of construction of short rules is NP-hard, so different heuristics are discussed in the literature. The paper presents comparison of two selected methods for decision rules construction. The first one is connected with a new algorithm based on EAV model, the second one - with construction of rules based on reduct. Decision rules were induced for data sets from UCI ML Repository and compared from the point of view of length and support, and from the point of view of classification accuracy. Results of Wilcoxon test are also included.

The paper presents research focused on decision reducts, a feature reduction mechanism inherent to rough sets theory. As a reduct enables to protect the discriminative properties of attributes with respect to described concepts, from the point of data representation, a reduct length is considered to be the most important measure of its quality. However, such approach is insufficient while taking into account the performance of a reduct-based rule classifier applied to test samples. When many reducts of the same length are available, they can lead to vastly different predictions. The paper provides a description for the proposed procedure for iterative reduct generation, which results in decrease of diversity in the observed levels of accuracy, supporting reduct selection. The procedure was applied for binary classification with balanced classes, for the stylometric task of authorship attribution.

The paper is dedicated to the area of feature selection, in particular a notion of attribute rankings that allow to estimate importance of variables. In the research presented for ranking construction a new weighting factor was defined, based on relative reducts. A reduct constitutes an embedded mechanism of feature selection, specific to rough set theory. The proposed factor takes into account the number of reducts in which a given attribute exists, as well as lengths of reducts. Two approaches for reduct generation were employed and compared, with search executed by a genetic algorithm. To validate the usefulness of the reduct-based rankings in the process of feature reduction, for gradually decreasing subsets of attributes, selected through rankings, sets of decision rules were induced in classical rough set approach. The performance of all rule classifiers was evaluated, and experimental results showed that the proposed rankings led to at least the same, or even increased classification accuracy for reduced sets of features than in the case of operating on the entire set of condition attributes. The experiments were performed on datasets from stylometry domain, with treating authorship attribution as a classification task, and stylometric descriptors as characteristic features defining writing styles.

The paper presents the proposed research methodology, dedicated to the application of greedy heuristics as a way of gathering information about available features. Discovered knowledge, represented in the form of generated decision rules, was employed to support feature selection and reduction process for induction of decision rules with classical rough set approach. Observations were executed over input data sets discretised by several methods. Experimental results show that elimination of less relevant attributes through the proposed methodology led to inferring rule sets with reduced cardinalities, while maintaining rule quality necessary for satisfactory classification.

The paper presents research on unsupervised and supervised discretisation of input data used in execution of stylometric tasks of authorship attribution. Basing on numeric characterisation of writing styles, recognition of authorship is performed by decision rules, as their transparent structure enhances understanding of discovered knowledge. The performance of rule classifiers, constructed in rough set approach, is studied in the context of a strategy employed for resolving conflicts. It is also contrasted with that of other selected inducers.

In this paper, seven greedy heuristics for construction of association rules are compared from the point of view of the length and coverage of constructed rules. The obtained rules are compared also with optimal ones constructed by dynamic programming algorithms. The average relative difference between length of rules constructed by the best heuristic and minimum length of rules is at most 4%. The same situation is with coverage.

In the paper, an application of dynamic programming approach for optimization of association rules from the point of view of knowledge representation is considered. The association rule set is optimized in two stages, first for minimum cardinality and then for minimum length of rules. Experimental results present cardinality of the set of association rules constructed for information system and lower bound on minimum possible cardinality of rule set based on the information obtained during algorithm work as well as obtained results for length.

In the paper, an application of dynamic programming approach to global optimization of approximate association rules relative to coverage and length is presented. It is an extension of the dynamic programming approach to optimization of decision rules to inconsistent tables. Experimental results wi th data sets from UCI Machine Learning Repository are included.

The paper describes a new tool for study relationships between length and coverage of exact decision rules. This tool is based on dynamic programming approach. We also present results of experiments with decision tables from UCI Machine Learning Repository.

In this work, we consider so-called nonredundant inhibitory rules, containing an expression “attribute:F value” on the right- hand side, for which the number of misclassifications is at most a threshold γ. We study a dynamic programming approach for description of the considered set of rules. This approach allows also the optimization of nonredundant inhibitory rules relative to the length and coverage. The aim of this paper is to investigate an additional possibility of optimization relative to the number of misclassifications. The results of experiments with decision tables from the UCI Machine Learning Repository show this additional optimization achieves a fewer misclassifications. Thus, the proposed optimization procedure is promising.

In this paper, we present three approaches for construction of decision rules for decision tables with many-valued decisions. We construct decision rules directly for rows of decision table, based on paths in decision tree, and based on attributes contained in a test (super-reduct). Experimental results for the data sets taken from UCI Machine Learning Repository, contain comparison of the maximum and the average length of rules for the mentioned approaches.

Based on dynamic programming approach we design algorithms for sequential optimization of exact and approximate decision rules relative to the length and coverage [3, 4]. In this paper, we use optimal rules to construct classifiers, and study two questions: (i) which rules are better from the point of view of classification – exact or approximate; and (ii) which order of optimization gives better results of classifier work: length, length+coverage, coverage, or coverage+length. Experimental results show that, on average, classifiers based on exact rules are better than classifiers based on approximate rules, and sequential optimization (length+coverage or coverage+length) is better than the ordinary optimization (length or coverage).

In the paper, we study a greedy algorithm for construction of decision trees. This algorithm is applicable to decision tables with many-valued decisions where each row is labeled with a set of decisions. For a given row, we should find a decision fro

In the paper, we study an extension of dynamic programming approach which allows optimization of approximate decision rules relative to the number of misclassifications. We introduce an uncertainty measure J(T) which is a difference between the number of rows in a decision table T and the number of rows with the most common decision for T. For a nonnegative real number γ, we consider γ-decision rules that localize rows in subtables of T with uncertainty at most γ.

This paper is devoted to the study of an extension of dynamic programming approach which allows optimization of partial decision rules relative to the length or coverage. We introduce an uncertainty measure J(T) which is the difference between number

The paper is devoted to the study of a greedy algorithm for construction of approximate tests (super-reducts). This algorithm is applicable to decision tables with many-valued decisions where each row is labeled with a set of decisions. For a given r

Partial association rules can be used for representation of knowledge, for inference in expert systems, for construction of classifiers, and for filling missing values of attributes. This paper is devoted to the study of approximate algorithms for minimization of partial association rule length. It is shown that under some natural assumptions on the class NP, a greedy algorithm is close to the best polynomial approximate algorithms for solving of this NP-hard problem. The paper contains various bounds on precision of the greedy algorithm, bounds on minimal length of rules based on an information obtained during the greedy algorithm work, and results of theoretical and experimental study of association rules for the most part of binary information systems.

This paper is devoted to the study of approximate algorithms for minimization of the total weight of attributes occurring in partial association rules. We consider mainly greedy algorithms with weights for construction of rules. The paper contains bounds on precision of these algorithms and bounds on the minimal weight of partial association rules based on an information obtained during the greedy algorithm run.