Abdelaziz Berrado

Supervised multivariate discretization in mixed data with Random Forests

Discretizing continuous attributes is necessary before association rules mining or using several inductive learning algorithms with a heterogeneous data space. This data preprocessing step should be carried out with a minimum information loss; that is the mutual information between attributes on the one hand and between attributes and the class labels on the other should not be destroyed. This paper introduces a novel supervised, global and dynamic discretization algorithm, called RFDisc (Random Forests Discretizer). It derives its ability in conserving the data properties from the Random Forests learning algorithm. RFDisc is simple, relatively fast and learns automatically the number of bins into which each continuous attribute is to be discretized. Empirical results indicate that the accuracies of classification algorithms such as CART when used with several data sets are comparable before and after discretization using RFDisc. Furthermore, C5.0 achieves the highest classification accuracy with data discretized with RFDisc when compared with other well known discretization algorithms.

Selection of Key Performance Indicators for Supply Chain monitoring using MCDA

Key Performance Indicators (KPIs) are very important for monitoring Supply Chains. The problem of selecting KPIs is considered as a multi-objective problem. This paper presents a framework for the ranking and the selection of KPIs using a Multi-Criteria Decision Analysis (MCDA). The research was carried out using Analytical Hierarchy Process (AHP) to make pairwise comparison of KPIs in terms of several criteria. The suggested framework is used for selecting relevant KPIs for monitoring a public sector pharmaceutical products supply chain in a developing country.

Towards a new framework for clustering in a mixed data space: Case of gasoline service stations segmentation in Morocco

Clustering is a widely used technique in data mining applications for discovering patterns in underlying data. Most traditional clustering algorithms are limited to handling datasets that contain either numeric or categorical attributes. However, data sets with mixed types of attributes are common in real life data mining applications. In this paper, we introduce a new framework for clustering mixed data which is based on Random Forest dissimilarity and PAM clustering. Then we apply this framework to segment market of services stations in Morocco to identify features that most influence on profit of each service station.

Class-association rules pruning using regularization

Association rules mining is a data mining technique that seeks interesting associations between attributes from massive high-dimensional categorical feature spaces. However, as the dimensionality gets higher, the data gets sparser which results in the discovery of a large number of association rules and makes it difficult to understand and to interpret. In this paper, we focus on a particular type of association rules namely Class-Association Rules (CARs) and we introduce a new approach of Class-Association Rules pruning based on Lasso regularization. In this approach we propose to take advantage of variable selection ability of Lasso regularization to prune less interesting rules. The experimental analysis shows that the introduced approach gives better results than CBA in term of number as well as the quality of the obtained rules after pruning.

Towards an interpretable Rules Ensemble algorithm for classification in a categorical data space

With the rapid growth of big data technology, classification plays an increasingly important role in decision making in many research areas. Several studies have been made in recent years to improve the accuracy-interpretability of classification models. In this paper, we present and discuss different classification methods, Random Forest, Boosting, CBA (Classification Based on Association) and Rulefit. We discuss the advantages and the limitations of each algorithm and finally we introduce a prototype model that combines some advantages that characterize the presented algorithms.

A new methodology for literature review analysis using association rules mining

Literature review is important for laying a strong foundation for scientific research. Summarizing previous studies, evaluating them and assessing their advantages, gaps and opportunities are key steps to progress in research. Multiple types of literature reviews have been developed and massively used in the past. This paper focuses on the quantitative literature reviews and reinforces available analysis methods with a new framework. The suggested methodology is based on association rules analysis. It brings a novel approach for analyzing data and exploring new research tracks by bringing to evidence the different relationships existing between the studied variables. An example is presented at the end of this paper to illustrate the framework.

Enhancing primary user detection through radio frequency fingerprint

Radio Frequency Fingerprint (RFF) is a technology that allows a unique identification of transmitters. RFF is based on the transient phase of a transmitted signal and allows device identification at the physical level. This paper proposes to use this technology to identify the primary user in the cognitive radio context. Indeed, it presents a novel transceiver architecture based on a dedicated sensing unit. Furthermore, we propose a decision making process based on a supervised learning classifier to decide if a given RFF belongs to a primary user or not. We use wavelets signal decomposition to extract RFF profiles in order to achieve a high level of sensing accuracy.