Browsing by Author "DJAFRI, Laouni"

Filter results by typing the first few letters
Now showing 1 - 2 of 2
  • No Thumbnail Available
    Item
    Imbalanced Datasets: Towards a better classification using boosting methods
    (University of Eloued جامعة الوادي, 2022-01-24) DJAFRI, Laouni; BACHA, Soufiane
    Imbalanced datasets classification is inherently difficult. This situation becomes a challenge when amounts of data are processed to extract knowledge because traditional learning models fail to generate required results due to imbalanced nature of data. In this paper, we will address the problem of imbalanced datasets whether at the class level, or at the classifier level. In our work, we are interested in binary or multi-class classification. To do this, we present a set of techniques used to solve this problem in particular boosting methods and machine learning algorithms. Our goal is therefore to re-balance the dataset at the class level and to find an optimal classifier to handle these datasets after balancing. Through the results obtained, it was observed that the boosting methods are well suited to re-balance the data and thus give a very satisfactory classification result.
  • No Thumbnail Available
    Item
    Machine Learning Algorithms for Big Data Mining Processing: A review
    (University of Eloued جامعة الوادي, 2022-01-24) DJAFRI, Laouni; GAFOUR, Yacine
    Big data mining is an excellent source of information and knowledge from systems to end users. However, managing such amounts of data or knowledge requires automation, which leads to serious consideration of the use of machine learning algorithms. Machine learning helps us make decisions if there is no right way to solve a problem identified in previous knowledge bases, and that is, too, one of the most widely used analysis and modeling tools for this purpose. In this work, we present an in-depth study that helps us to choose the best machine learning algorithms in order to process big data and extract knowledge from it, so that, this treatment can be very flexible, either in a simple system with sequential computing, or in a distributed system with parallel computing. To achieve this, we will, first and foremost, test the accuracy of the results provided by the classifiers; here we mean the strength and flexibility of a classifier when it comes to dealing with big data mining. Second, we will also test the execution speed for each classifier in complex cases; that is, when the classifier will not be sufficient to solve a particular problem in the context of big data mining, especially if all cases are dealt with quickly and efficiently. The results obtained in this paper demonstrated the superiority of certain classifiers over others in certain cases, and demonstrated their failure in other cases, the reason being due to the nature of the dataset, in particular the number of instances, the number of attributes , and the number of classes.