As part of the 4th and 5th-year industrial innovation projects included in the engineering curriculum, sixteen ESILV students have set themselves an ambitious challenge: to simplify banking, actuarial and green finance tools as much as possible, to offer partner companies and their users products that will make their lives easier, safer and more practical.
Industrial innovation projects for engineering students are moments of real self-surpassing and privileged moments for Pole’s de Vinci corporate partners to find inspiration in students designs and creations to recurring problems in finance, thanks to project-focused group work.
Credit Scoring : a Geolocation Tool
Nathan BIBRAC – Nicolas DIRMANN– Nima FAGANDAHR – Pierre GROUARD
Fourth and fifth-year students created a geographical data-based credit scoring tool, to help banks and credit issuers.
“We used machine learning algorithms to establish this credit score. The latter is a function that assigns a credit quality value to a customer or a loan based on explanatory variables such as the borrower’s debt-to-income ratio, account behaviour or any other magnitude that is correlated with borrowers who are in default.”
The scoring tool that links the explanatory variables to the magnitude of loan defaults is then refined using a training set and a validation sample. The data thus obtained is verified and validated according to a training set that confirms the relevance of the selected model.
A simplified database for actuaries
Kévin AUDOUY – Maxime CAMPANE – Anya HAOUCHINE – Imanne M GHAITE
“No more paper reports, no more endless days of searching through mountains of data! Imagine an algorithm designed for you, insurer. Imagine being able to clean data, detect anomalies, reduce dimensions, create groups of individuals to be able to easily classify your clients”.
The group used the R language to make a 3-step algorithm :
- Dataset (database) exploration: data selection, calibration and replacement;
- Detecting and removing anomalous data through tools such as k-Means, k-Nearest-Neighbours (kNN) and Local Outlier Factor (LOF), which determine the degree of normality and aberrance
- Display of groups of individuals via clustering methods (kMeans, ACH, WFP, FCM-Fuzzy, etc.)
These methods, combined with a generic and adapted code, enabled ESILV students to address issues encountered by Allianz actuaries. The project team developed an optimized, efficient and user-friendly tool that will allow Allianz Group’s actuarial teams to have a clear and complete vision of potential clients.
Algorithmic Transparency Solution
Audrey ESSEUL – Fabien KFOURY – Emily RAJIBAN – Kirthikaa THAYANITHY
The four 5th year students launched a data challenge to make the algorithms transparent as part of an evaluation and comparison program inspired by improving Societe Generale‘s group Machine Learning process.
To do this, they retrieved datasets adapted to Machine Learning from Kaggle – the largest community of data scientists in the world!
“We created an R Shiny interface, an RStudio library, allowing us to highlight interesting features, such as the distribution of variables, tables and confusion matrices in order to visualize the two models and to know which one was better.”
The two compared algorithms were: a linear algorithm, logistic regression and a non-linear algorithm, Random Forest.
“Finally, the contact with Société Générale brings a professional aspect to the project, by making it complete and enriching. In addition, through this project, we had the opportunity to develop skills and know-how about algoritms transparence, that will be useful to us throughout our professional career.”
Big data for Green Finance
Paul CHARTIER – Violette LEFRANC – Miguel Angel RAMOS FERRUFINO – Maxime TAZI
As part of their partnership with Dim Tech, a global Sino-French investment management and data science company, the 4 5th year students developed an artificial intelligence algorithm that detects the long-term investment opportunities in the Cleantech.
This subject is quite interesting because, currently, a lack of information makes it hard to provide a forecast on investment for the startups offering green solutions. Therefore, the students proposed an engineering solution that would allow investment specialists to collect and analyze website data such as scientific articles, economic reports. In other words: Applying the Big Data methodology used in ﬁnance to forecast investment returns in the green sector.
The students started from the market finance-based methods to explore green advancements by comparing data from startups’ websites to data extracted from the scientific opinion. More precisely, the ESILV team retrieved data, such as keywords, computerized information, from several articles published in various scientific journals on recent green technologies.
The process allowed students to set scientific criteria to identify innovation indicators, which will be then used by DimTech group to analyze viability of green tech projects.
“The ideal way is to automate the whole process, however, we quickly find out the difficulty of doing so. Indeed, creating a generalized web scraping system is a very important challenge lead to an accuracy drop and a risk of error propagation.”