The urinary system is the organ system responsible for the production, storage and elimination of urine. This system includes kidneys, bladder, ureters and urethra. It represents the major system which filters the blood and any imbalance of this organ can increases the rate of being infected with diseases. The aim of this paper is to evaluate the performance of different variants of Support Vector Machines and k-Nearest Neighbor with different distances and try to achieve a satisfactory rate of diagnosis (infected or non-infected urinary system). We consider both diseases that affect the urinary system: inflammation of urinary bladder and nephritis of renal pelvis origin. Our experimentation will be conducted on the database “Acute Inflammations Data Set” obtained from UCI Machine Learning Repository. We use the following measures to evaluate the results: classification accuracy rate, classification time, sensitivity, specificity, positive and negative predictive values.

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.

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.