Cancer diagnosis is one of the most studied problems in the medical domain. Several researchers have focused in order to improve performance and achieve to obtain satisfactory results. Breast cancer is one of cancer killer in the world. The diagnosis of this cancer is a big problem in cancer diagnosis researches. In artificial intelligent, machine learning is a discipline which allows to the machine to evolve through a process. Machine learning is widely used in bio informatics and particularly in breast cancer diagnosis. One of the most popular methods is K-nearest neighbors (K-NN) which is a supervised learning method. Using the K-NN in medical diagnosis is very interesting. The quality of the results depends largely on the distance and the value of the parameter "k" which represent the number of the nearest neighbors. In this paper, we study and evaluate the performance of different distances that can be used in the K-NN algorithm. Also, we analyze this distance by using different values of the parameter "k" and by using several rules of classification (the rule used to decide how to classify a sample). Our work will be performed on the WBCD database (Wisconsin Breast Cancer Database) obtained by the university of Wisconsin Hospital.

No abstract

No abstract

Feature extraction is an important step in image classification. It allows to represent the content of images as perfectly as possible. However, in this paper, we present a comparison protocol of several feature extraction techniques under different classifiers. We evaluate the performance of feature extraction techniques in the context of image classification and we use both binary and multiclass classifications. The analyses of performance are conducted in term of: classification accuracy rate, recall, precision, f-measure and other evaluation measures. The aim of this research is to show the relevant feature extraction technique that improves the classification accuracy rate and provides the most implicit classification data. We analyze the models obtained by each feature extraction method under each classifier.

No abstract

Stochastic optimization is widely used in combinatorial computer vision problems, and many variant have been proposed. This contribution aims at analyzing and assessing several flavours of the simulated annealing algorithm. We particularly want to show the optimization performance, convergence speed, and quality of the solution with respect to the algorithm's parameters and cooling schedules. We also verify experimentally that the S.A. algorithm is a global method i.e. it is able to lock a strong minimum regardless of the initialization. Performance evaluation is conducted in the context of stereo matching.

Abstract-In this paper, feature selection and parameters determination in SVM are cast as an energy minimization procedure. The problem of feature selection and parameters determination is a very difficult problem where the number of feature is very large and where the features are highly correlated. We define the problem of feature selection and parameters determination in SVM as a combinatorial problem and we use a stochastic method that, theoretically, guarantees to reach the global optimum. Several public datasets are employed to evaluate the performance of our approach. Also, we propose to use the DNA Microarray Datasets which are characterized by the large number of features. To validate our approach, we apply it to image classification. The feature descriptors of the images were extracted by using the Pyramid Histogram of Oriented Gradients. The proposed approach was compared with twenty feature selection methods. Experimental results indicate that the classification accuracy rates of the proposed approach exceed those of other approaches.

No abstract