Marwen S.

Scoring for French companies based on their financial status • Extraction, cleaning, preparation and pre-processing of raw data hosted by Infogreffe • Feature engineering in order to build learning database • Development of a classification machine learning model that predicts the probability of companies default based on their annual balance sheets & compute the scores • Development of an interpretation module that explain the output of the machine learning module - Design and implementation of an end-to-end pipeline that execute the whole processing (from extraction to score calculation) Keywords : Probability of failure, Accounting and Finance, Dataiku DSS, machine learning, Postgres

• Machine Learning: Development and validation of custom machine learning prediction models for the estimation of some power plant's subsystem output. (Supervised Learning, Deep Learning, Box & Jenkins Models, Cross Validation, Python). • Statistical causality analysis: Estimation of the causality relationships between the various sensor signals (Lasso-Granger Causality, Graph Theory, Python) • Feature Selection tool: Implementation of an algorithmic system for the elimination of redundant signals to maximize the information. (Correlation analysis, Linear & non-linear dependency, Python) • Visualization tool: Development of a visualization tool to analyse the interactions between the different signals of the power plant. (Dimensionality reductions, Interactive visualization, Python) • Data Extraction: Development of a scalable script for the extraction of sensor signals from the log files of a power plant’s archives. (Java DataBase Connectivity, SQL, Design Pattern, NoSQL Database)

• Software Engineering: Implementation of an IT solution that allows customers to use their inbuild statistical models developed in the commercialized software of Mass Analytics. (Java, R, Renjin (API between R et Java), Design Pattern) • Research and Development: Investigate the use of non-linear models and evaluate their modelling capabalities and their relevance in interpretation compared to traditional linear models. (Log-linear model, Bayesian networks, R) • Project management: Teamwork on programming tasks through software versioning and revision control using agile methods. (Agile method Scrum, Apache SVN)

• Forward kinematics and Inverse Kinematics formulation for the in-house developed humanoid robot CyHRoP (Matlab) • Trajectory planing for the center of mass using Linear Inverse Pendulum method (Matlab) • Implementation of the developed algorithms into the robot (Simulink, Arduino)