Yazar "Ata, Sezin" seçeneğine göre listele

Listeleniyor 1 - 2 / 2
  • Küçük Resim
    Yayın
    Searching for the optimal ordering of classes in rule induction
    (IEEE, 2012-11-15) Ata, Sezin; Yıldız, Olcay Taner
    Rule induction algorithms such as Ripper, solve a K > 2 class problem by converting it into a sequence of K - 1 two-class problems. As a usual heuristic, the classes are fed into the algorithm in the order of increasing prior probabilities. In this paper, we propose two algorithms to improve this heuristic. The first algorithm starts with the ordering the heuristic provides and searches for better orderings by swapping consecutive classes. The second algorithm transforms the ordering search problem into an optimization problem and uses the solution of the optimization problem to extract the optimal ordering. We compared our algorithms with the original Ripper on 8 datasets from UCI repository [2]. Simulation results show that our algorithms produce rulesets that are significantly better than those produced by Ripper proper.
  • Küçük Resim
    Yayın
    Searching for the optimal ordering of classes in rule induction
    (Işık Üniversitesi, 2012-09-19) Ata, Sezin; Yıldız, Olcay Taner; Işık Üniversitesi, Fen Bilimleri Enstitüsü, Bilgisayar Mühendisliği Yüksek Lisans Programı
    In this thesis, we work on rule induction algorithms, basically Ripper. These algorithms solve a K>2 class problem by transforming it into a sequence ok K-1 two class problems. As a heuristic, these algorithms learn classes in the order of increasing prior probabilities. Although the heuristic works well in practice, there is much room for improvement. We propose two algorithms for that purpose. The first algorithm, namely Forward Ordering Search(FOS) starts with the ordering heuristic provided and searches for better oderings by swapping consecutive classes. For a dataset with K classes, the ordering space will be as large as K!. Since FOS is an example of Steepest Ascent Hill Climbing(Gradient Search), starting with the heuristic ordering will only give local maximum in the search space. In order to improve the performance, we use 10 random initial orderings as in Random-Restart (Steepest Ascent) Hill-Climbing. The best performance between 10 random initial orderings is the result of Random-Restart FOS. The second algorithm, namely Pairwise error Approximation (PEA), transforms the ordering search problem into an optimization problem and uses the solution of the optimization algorithm to extract the optimal ordering. In this algorithm, the number of random orderings to construct the optimization problem is a parameter and we try several values of this parameter to see the effect on the performance. We compare our algorithms with the original Ripper on 13 datasets from UCI repository [1]. Experimental results show that, our algorithms produce rule sets that are significantly better than those produced by Ripper proper in general and the number of rules and conditions of the produces rule sets are comparable with Ripper proper. Even though the accuracy of Random-Restart FOS is better than FOS, the time complexity of the algorithm is far worse than FOS. The average error estimation results of PEA promote the consistency of our pairwise assumption and show the relationship between accuracy and the number of random orderings to extract the optimal ordering.