RINA - Démonstrateur d’utilisation de l’intelligence artificielle pour la gestion opérationnelle des risques naturels gravitaires géologiques | FEREC 2020

0.1. INTRODUCTION 5 0.1 Introduction In this report, we study rockfall characteristics in Réunion RN1 road and aim at building a pre- dictive model, which could replace existing, empirical, operational rules. Understanding more precisely dynamics and dependencies of the process will allow us to design relevant data-driven model. This work have been done by LISTIC as part of RINA project, asked by the other part- ner(CEREMA, BRGM, Géolithe), which are more in the applied fields of rockfall. Besides, a complementary work on other machine Learning models (decision trees, nearest neighbour) on same data has been achieved by CEREMA trough the internship of Guilherme Cunha de Barros Santos. LISTIC follow up this second work and supervise it on the machine learning part. In Chapter 1, we present available data, their format, first analysis with previous work and descriptive statistics about on the dynamic of rockfall event. In Chapter 2, we precise different mathematical manner to represent rockfall events. Some relevant papers , linked to them, are presented in Annexe B. Chapter 3 summarize numerical experiments with different models (Neural Networks, General linear Model...), comparing best models with expert rules used in operational center. In conclusion, we provide some preliminary results on the capability of machine learning to predict rockfall and some perspectives of improvement.