Abstract:

In a context of an online collaborative course, students worked in groups in order to perform a series of activities. The aim of the course is to achieve a certain level of skills in group work. Predictive modeling is an important part of learning analytics; we used it in this research in order to estimate the student skills’ level after accomplishing the whole course with its activities. Data is collected by using a communication activity in class followed by an online questionnaire. A total of 119 students during 4 consecutive years, following the same course, is included in the study. We used Random Decision Forests machine learning algorithm, as a method for classification of students, with Microsoft Azure Machine Learning (MAML) as a tool for designing the model and evaluating its performance and accuracy. Preliminary results presented high level of accuracy of the predicting model.

Introduction:
In collaborative learning, knowledge is defined as a process of joint construction in which interaction between peers is basic, although it also involves the teacher. It is not a question of the circumstantial application of group techniques (more typical of cooperative learning), but of promoting the exchange and participation of all members in the construction of knowledge. An important part of learning analytics is Predictive modeling for teaching and learning. For this purpose, computational techniques are involved and wide varieties of methods have been applied for predicting students' performance in general. Nevertheless, the use of one technique depends on many criteria related to the subject of prediction; such as the data number or the data category; label or quantity used in the prediction model.

Online Collaborative Course:
Within the framework of a course named “Tools of collaborative work”, we wanted to reflect, in the scenario of the series of works to be performed, activities of a team in a company. Therefore, it has essentially two main objectives, the learning of the remote group work and the learning of the use of the maximum of ICT tools. Moreover, an online space is reserved for this course and an online base of free tools is put at the disposal of the students, containing all the tools to be used. This course is spread out on a half-year and is given to students of Information technology specialty.

Predicting model:
In our case, we used the Random Decision Forests algorithm in a machine learning modeling in order to predict student skills of group work. In order to adjust the model and save time, we used MAML, which allowed us to design, test and evaluate the model. Our problem is represented by the fact that not all students who have taken the same course have acquired the required skills. The Scikit-learn algorithm helped us define the appropriate machine learning algorithm for the predictive modeling which is Classification. Consequently, the model would predict to which class a student belongs. We had to decide which algorithm to use. Our choice has gone to the Random Decision Forest, which is a variety of Decision Tree Algorithm.

Preliminary Results:
As preliminary results, the model gave 0.821 of accuracy, which presents a good evaluation level of prediction. We are playing with the different parameters in order to improve more the model. Otherwise, we are working on predicting two other skill acquisition, which are the ICT use and interculturality.

Keywords:

Student skills, Online Collaborative Learning, Predictive Modeling, Classification, machine learning.