Analysis of training accompaniment needs through prediction models assisted by machine learning

The purpose of this study is to analyze tendencies in the needs of students in accompaniment in a perspective of prediction of the measures to be taken during the training. This approach consists in measuring, with adapted prediction models, the tendencies of accompaniment needs in three areas of competence of the training: competencies practices, written competencies and oral competencies. To this end, the accuracy of the models in these three areas of competence must be verified in order to classify their prediction parameters. In a first step we used data modeling of machine learning with data partitioning, 70% learning, 30% testing of all data. Then we compared the predictive models (SVM, Neural Network, Bayasian Network, CART, CHAID, C5) using the global precision index. This allowed us to select the best model based on its accuracy performance in the three areas of expertise already mentioned. Three models were selected among the six explored for their levels of accuracy in the three areas of expertise. The accuracy performance of the prediction models is therefore distributed as follows based on the domains: Practical skills (Neuronal Network) with an accuracy percentage of 68.66%; Written skills (CART) with an accuracy percentage of 95.184%; Oral skills (Bayesian Network) with an accuracy percentage of 66.6%. Prediction fields are defined from these analyses, containing a set of statements delineating the use of 7 predictors for the domain of practical competencies, 8 predictors for the domain of written competencies and finally 7 predictors for the domain of oral competencies. These determined predictors are therefore the assets for our decisions regarding the design and implementation of possible support systems for the regulation of training.