Prediction model

Initial situation

The teacher’s intention is to allow students to approach the concept of “Machine Learning” in an applied way. The goal is that the work done in class be practical and representative of the active world. To do so, the teacher has created an internal data mining competition for the students. The exercise is carried out on real data from companies that have put them online.

Project implementation

The course starts with 2 to 3 sessions during which the concept studied as well as the theoretical and practical bases are defined. The teacher then proposes to take part in the competition using the Kaggle platform. This platform allows many companies to make available large databases to meet a specific demand. The students are thus placed in an intermediate situation between course and reality.

During the next 8 weeks, the students participate in groups of 2 in this competition. Using a sample of the database provided, their objective is to develop the best prediction model (“machine learning model”) of the behavior of the rest of the data. To do this, groups can submit up to one model proposal per day if they wish. For each proposal submitted, feedback is sent allowing them to improve their model while working independently. The frequency with which they participate in this competition is not imposed, which allows them to take responsibility.

At the same time, the students participate in practical work organized by the assistants. During these practical work sessions, they carry out practical exercises and can also ask questions related to the models developed during the competition.

Part of the evaluation (40%) is the written report at the end of the semester in which the students present their thoughts on the proposed model. The models are applied to the whole database and the 3 models that best predict the data are rewarded with a bonus for the final exam. The final exam is in a classical written form and counts for 60% of the final grade.

Thoughts and advice

One of the difficulties during the implementation of this project concerns the choice of the exploited database. This one should be neither too simple nor too complex. The risk is to demotivate or discourage the students.

The implementation of such a project brought a lot to its creator and allowed him to rethink his way of teaching by getting more directly involved in his teaching by interacting more with the students.

On the one hand, the very applied aspect and the notion of competition are characteristics that motivate students to take part in the course and to push themselves. On the other hand, they sometimes tend to underestimate the amount of time to invest outside of class and may feel overwhelmed.

Student feedback

“The course is complete, with a theoretical side, a practical side with the report, and a competition side with the challenge on Kaggle.” “The course is very much application-oriented. The structure is also clear. The data competition is a very nice way to make the learning process more dynamic.” “The data competition is interesting and challenging.” “The data competition may redirect focus from the actual objectives of the course.” “Very good course! Demanding content, but teacher stimulates reflection and the Kaggle competition is a good incentive for us to do some real ML!”