The main objective of this study was to implement a system capable of recognition and identification of students of the Université de l’Assomption/Institut Supérieur de Développement de l’Assomption (UAC/ISDA) exams’ rooms. The system will serve as a means to fight against fraud during exam periods, i.e. before a student enters the exam room, the system will identify the student by displaying from his/her captured face the full name and the amount of fees already paid by the student whose face appears on video sequences from Smartphone cameras. This study used 2210 images of human faces at a rate of 10 images per class to train the deep CNN model. Each class was labeled with the identifiers of the respective student such as first name, surname name, promotion and the amount of academic fees already paid. After training the neural network, the auteurs obtained an accuracy of 94% and a loss of 0.25 as validation results. The test was performed on 30 images captured in four different locations of which 29 predicted correctly or 96.66%. These results showed the effectiveness of the artificial neural network for the recognition and identification of UAC/ISDA students. This model answers our problem in the sense that it gives the possibility to identify not the object but the person.

Several data mining techniques are used to extract hidden knowledge in educational data to help students make a useful decision for their university orientation. Indeed, every year, students are enrolling in universities, the massive arrival of these candidates poses the thorny problem of orientation. The hidden problem behind this orientation is the lack of information concerning the possibilities of orientation; or the lack of support from the close entourage. Having developed the survey questionnaire, the authors collected 712 responses. After analyzing these data, they trained the models and measured their performance with four evaluation measures: accuracy, precision, recall and the F-score. The results of these models showed that the SVM algorithm gave 70% accuracy, the Naïve Bayes 65% Accuracy, the Neural Network 64% and the decision tree gave only 52%. This allowed SVM to be selected as the model that predicted better than the others. Finally, the authors deployed the validated model in web technology using Flask.