Résumé: This paper investigates feature selection methods based on hybrid architecture using feature selection algorithm called Adapted Fast Correlation Based Feature selection and Support Vector Machine Recursive Feature Elimination (AFCBF-SVMRFE). The AFCBF-SVMRFE has three stages and composed of SVMRFE embedded method with Correlation based Features Selection. The first stage is the relevance analysis, the second one is a redundancy analysis, and the third stage is a performance evaluation and features restoration stage. Experiments show that the proposed method tested on different classifiers: Support Vector Machine SVM and K nearest neighbors KNN provide a best accuracy on various dataset. The SVM classifier outperforms KNN classifier on these data. The AFCBF-SVMRFE outperforms FCBF multivariate filter, SVMRFE, Particle swarm optimization PSO and Artificial bees colony ABC.

Résumé: Automatic speaker verification (ASV) is a binary classification task. It consists of accepting or rejecting the claimed identity. ASV system has to decide whether a claimed speaker uttered a sentence. This paper proposes an algorithm called impostor random vector quantisation (IRVQ) based on multiples random codebook. IRVQ represents the impostor model also called universal background model (UBM) and we compare it to the second algorithm called partial impostor VQ (IVQ) for vector quantisation (VQ) in speaker verification. The present study demonstrates that the several random selected codebooks representing impostor models give better results and less half total error HTER than impostor IVQ method and baseline system. The performance of these models is evaluated on Arabic speaker verification dataset. However, this improvement also depends on the codebook size. The assumption concerning partitioning impostor acoustic spaces and choosing the best subspace to represent impostor model specific to each speaker is an efficient approach.

Résumé: The performance of speaker verification system degrades when the test segments are utterances of short duration, therefore, we investigate the use of model representing our target speaker with his close speaker and his own speech data. We propose to create a new Speaker Model who groups close speakers (CS) achieved with two clustering algorithms in Automatic Speaker Verification A.S.V. Intra and Inter speaker’s variability are two clustering algorithm used in voice module. We compare the traditional approach which uses one specific customer model (Maximum a Posteriori Adaptation) with the Close Speaker model (Customers Families).Close Speaker Model (CSM) applied only when speaker model is weak achieves 42% of equal error rate. The results demonstrate that the log likelihood of close speakers is greater than the likelihood of client speaker. The false alarm from client and CSM are closest and we are constrained to enhance speaker model.

Résumé: This paper investigates Chaotic Restart Binary Particle Swarm PSO (ChResBPSO) algorithm applied to feature selection. In this study, to escape local optima of particle swarm PSO and solve the stagnation problem, we add new particles using prior information about current global best and its neighborhood. The solution adopted is to update the particles using N-Best previous particles, their neighbors and new random particles. The information on worst features is incorporated to direct the novel solutions to avoid them. Various chaotic systems replace the main parameters of PSO to find the best chaotic map. Experiments conducted on UCI data: hepatitis, breast cancer, colon cancer, DLBCL validate that chaotic PSO with anti-stagnation criterion outperforms the state of the art methods (BPSO), chaotic BPSO, artificial bee colony (ABC). The novel ChRes-BPSO enhances the final solution in term of accuracy and minimal number of features.

Résumé: This article proposes a Novel Search Equation of Artificial Bee Colony called NSABC. This study introduces new search equations for employed and onlooker bees involving more parameters for each food source (more features) and avoid local optima. NSABC outperforms various ABC variants and particle swarm optimization PSO in term of reduced size of feature and accuracy. Experimental results validate the efficiency of NSABC method on machine learning UCI data.

Résumé: This paper investigates feature selection method using two hybrid approaches based on artificial Bee colony ABC with Particle Swarm PSO algorithm (ABC-PSO) and ABC with genetic algorithm (ABC-GA). To achieve balance between exploration and exploitation a novel improvement is integrated in ABC algorithm. In this work, particle swarm PSO contribute in ABC during employed bees, and GA mutation operators are applied in Onlooker phase and Scout phase. It has been found that the proposed method hybrid ABC-GA method is competitive than exiting methods (GA, PSO, ABC) for finding minimal number of features and classifying WDBC, colon, hepatitis, DLBCL, lung cancer dataset. Experimental results are carried out on UCI data repository and show the effectiveness of mutation operators in term of accuracy and particle swarm for less size of features.

Résumé: This paper investigates feature selection stage applied to medical classification of disease on datasets from UCI repository. Feature selection methods based on minimum Redundancy Maximum Relevance (mRMR) filter and Ficher score were applied, each of them select a subset of features then the selection criteria is used to get the initial features subset. The second stage Support vector machine recursive feature elimination is performed to have the final subset. Experiments show that the proposed method provide an accuracy of 99.89 % on hepatitis dataset and 97.81 % on Wisconcin Breast cancer dataset and outperforms MRMR and Support vector machine recursive feature elimination SVM-RFE methods, as well as other popular methods on UCI database, and select features that are relevant in discriminating cancer class (malign/benign).

Résumé: This paper presents a work consisting in realizing a decision support system based on the technique of case-base reasoning and dedicated to the diagnosis of a very dangerous pulmonary pathology: lung cancer. The system is realized for the oncology department of Ibn roch hospital of Annaba (Algeria) and will help young oncologist physicians in their activity by providing them with the experience of experts in the same domain. The principle issue in this work is the missing data in the system memory relating to the patient’s state. Indeed, missing values prevent the achievement of the diagnosis process. The problem is treated by proposing two statistical approaches in addition to re-evaluate in this new domain some ones which have been already proposed and evaluated in a previously domain. The validation is made on a base of 40 real cases collected from the archive of oncology department. Cases are collected as paper documents.

Résumé: This paper investigates initializations on both Chaotic Particle Swarm PSO and reduced PSO algorithm. The first initialization method is chaotic initialization with three variants of chaotic PSO. Several chaotic maps are tested. Moreover, to achieve improvement for PSO algorithm, the second initialization method called reduced size PSO (RedPSO) is characterized with limited size of features depending on the whole size of features. It has been found that the chaotic initialization method is better than random initialization for exiting methods (PSO), artificial bee colony (ABC) in term of minimal number of features and highest accuracy. Experimental results validate the chaotic initialization approach tested on UCI data.

Résumé: This paper investigates feature selection method using filter Fast Correlation based Filter FCBF combined with Genetic Algorithm GA and particle swarm optimization PSO. In this paper two hybrid approaches based on filter method FCBF and Genetic algorithm (FCBF-GA) and filter FCBF with particle swarm (FCBF-PSO) are proposed. It has been found that the proposed method FCBF-PSO outperform the proposed FCBF-GA method and exiting methods (FCBF, GA, PSO) for classifying WDBC, colon, hepatitis, DLBCL, lung cancer dataset. Experimental results are carried out on UCI data repository and show the effectiveness of these approaches in term of accuracy and reducing the size of features.

Résumé: This paper investigates image segmentation applied to mammographic images and deals with image segmentation using two variants of spatial clustering algorithm. The first algorithm is based on spatial Fuzzy c-means SFCM using mean filtering called SFCM_S1 and the second also founded on spatial fuzzy c-means but applying median filtering called SFCM_S2. We propose to use SFCM_S1 and SFCM_S2 to segment the breast then, apply cluster Validity as metrics to obtain the best number of regions. The cluster validity functions evaluated are Partition Entropy (PE) and Partition Coefficient (PC) to get the best partitions and the number of clusters. The MIAS database was used in the experiments to validate this approach. The results indicate that this architecture is a good approach better than fuzzy c-means but additional cluster validity indexes have to be tested.

Résumé: Glaucoma is a group of eye diseases that cause an irreparable decrease in the view field which is the second most common and leading causes of blindness among retinal diseases. Computer aided diagnosis (CAD) system is an emerging field in medical informatics which has high importance for providing prognosis of diseases. Research efforts have reported with increasing confirmation that the twin support vector machines (TWSVM) have greater accurate diagnosis ability. The goal of TWSVM is to construct two non-parallel planes for each class such that each hyper-plane is closer to one of two classes and as far as possible from the other one. In this paper, we propose a new CAD system for glaucoma diagnosis using TWSVM and three heterogeneous families of feature extraction. In this work, we have used 169 images to classify into normal and glaucoma classes. The performance of the method is evaluated using classification accuracy, sensitivity, specificity and receiver operating characteristic (ROC) curves. The results show that the highest classification accuracy (98,53%) is obtained for the TWSVM using Gaussian kernel function, and this is very promising compared to the SVM results.

Résumé: This paper investigates feature selection stage applied to medical classification of disease on datasets from UCI repository. Feature selection methods based on minimum Redundancy Maximum Relevance (mRMR) filter and Ficher score were applied, each of them select a subset of features then the selection criteria is used to get the initial features subset. The second stage Support vector machine recursive feature elimination is performed to have the final subset. Experiments show that the proposed method provide an accuracy of 99.89 % on hepatitis dataset and 97.81 % on Wisconcin Breast cancer dataset and outperforms MRMR and Support vector machine recursive feature elimination SVM-RFE methods, as well as other popular methods on UCI database, and select features that are relevant in discriminating cancer class (malign/benign).

Résumé: This article investigates the vector quantization approach applied to speaker identification (SID) in two languages Arabic and French. We expect to recognize speaker even whether the training process was done only in one language and the test is achieved with another different language. The database used is created with one hundred speakers in Arabic and French. The results indicate that the performance of SID system is severely degraded when different languages are used.

Résumé: This paper explores the use of mismatch condition in speaker variability applied to Automatic Speaker Verification (ASV) defined as a classification task to decide whether a proclaimed identity is true or not. This paper proposes to model mismatch conditions in speaker variability from session to another. It was shown that the speaker recognition accuracy deteriorates when there is an acoustic mismatch between the speech obtained during training and testing. The target speakers are modeled using vector quantization (VQ) approach based on multiples session codebooks, and then compared it to baseline vector quantization. The present study demonstrates that the session variability for target speaker improve the performance of ASV system and the several codebooks for target speaker (set of session codebooks) give better results.. The performance of these models is evaluated on the Arabic speaker verification dataset.

Résumé: L’objectif de ce travail est de construire un classifieur Bayésien Naïf et un arbre de décision qui aide à la classification des tumeurs du sien en utilisant des symptômes cliniques et des caractéristiques extraites de l’image mammographique. Notons que l’approche utilisée dans ce travail est une approche statistique de classification. La première approche est basée sur un modèle probabiliste dérivant le théorème de Bayes. Nous proposons un classifieur de bayes naïf capable de classifier le type du cancer du sien sur une mammographie. Dans la même thématique, nous traitons dans ce travail la classification automatique par la méthode d’arbre de décision. La classification des tumeurs a été faite par l’algorithme ID3 (Inductive DecisionTree). Nous avons validé les deux classifieurs sur une base réelle des images mammographiques (MIAS), les résultats obtenus sont satisfaisants.

Résumé: Speaker Verification (ASV) is a binary classification task to decide whether a claimed speaker uttered sentences. This paper proposes two different algorithms for vector quantization (VQ) to speaker verification. The first algorithm named Partial Vector Quantization (Partial VQ) is based on partitioning acoustics space, represents the impostors called universal background model(UBM) and compared it to second vector quantization algorithm Reduced UBM session used for keeping the relevant training data. The present study demonstrates that several codebooks for Universal Background Models give better results. The performance of these models is evaluated on the Arabic speaker verification dataset. The VQ Partial method achieved less half total error rate for 128 codebook size better than Baseline Vector Quantization approach for 32 codebook sizes.

Résumé: This article investigates several technique based on vector quantization (VQ) and maximum a posteriori adaptation (MAP) in Automatic Speaker Verification ASV. We propose to create multiple codebooks of Universal Background Model UBM by Vector Quantization and compare them with traditional approach in VQ, MAP adaptation and Gaussian Mixture Models.