Temporary line plot placeholder

# Machine Learning Approaches for the Prediction of Credit Risk G. Ausset - [theses.fr](http://www.theses.fr/s197391) - [pdf](/thesis/thesis.pdf) - [slides](/thesis/slides.pdf) ## Abstract Predicting the possible occurrence of a future event, which may eventually never happen, is a fundamental problem that naturally occurs in most scientific as well as industrial fields. This problem, commonly referred to as survival analysis after its canonical application in epidemiology, has long been one of the classical problems in statistics whose exceptional contributions have enabled immeasurable advancements in the natural sciences. More recently, through advancements in the field of machine learning, those same natural scientific fields and industrial applications have also been able to achieve significant leap forwards by exploiting large amounts of high-dimensional data using highly flexible estimators. In this thesis we try to reconcile both approaches and show how to best make use of the highly flexible machine learning approaches in the survival analysis setting in a principled and motivated way. We show in this work how the classical ERM framework can be adapted to the survival analysis setting by introducing a reweighted objective called the Kaplan-Meier ERM and derive non-asymptotic error bounds without parametric assumptions on the true generating process, effectively bringing the results one has come to expect in the machine learning field to survival analysis. We also show how to construct highly flexible estimators of the survival function, one of the key building blocks of our Kaplan-Meier ERM framework. We formulate the survival as a normalizing flow problem and introduce a novel conditional normalizing flow estimator of the survival density, giving a tractable, easy to sample from, but highly expressive estimator of the survival density. In order to reduce the complexity of the two previous approaches, we introduce an estimator of the gradient of a black box function and show how to use it for variable selection, a simple yet highly effective method for dimensionality reduction. Finally, we apply the methods developed here to a particular instance of the survival problem: predicting the defaults of companies. We show how to use estimators of the probability of default to build optimal portfolios as well as how to efficiently make use of small data through hierarchical methods. ## Details This is temporary layout filler for checking the paper-page template. Replace it with human-written notes before publishing. ## Normal content heading This paragraph exists to test the rhythm of ordinary prose inside the extended paper body. It should read like normal content, unlike the fixed `Abstract`, `Details`, and `BibTeX` labels around it. Inline math should work in running text, for example $\hat f_n(x)$, $S(t \mid x)$, and $\mathbb{E}[Y \mid X=x]$. The display math below is only here to test spacing and overflow behavior. $$ \widehat{R}_n(f) = \frac{1}{n}\sum_{i=1}^{n} \frac{\delta_i}{\widehat{G}(T_i \mid X_i)} \mathcal{L}\bigl(T_i, f(X_i)\bigr) $$

Temporary figure caption for testing line length, spacing, and caption styling.

## Tables, lists, and small blocks | Element | Purpose | Notes | |---|---:|---| | Inline math | spacing | $x_i \in \mathbb{R}^d$ | | Display math | overflow | long equations should not break the page | | Figures | captions | captions should stay visually secondary | - A short bullet item. - A longer bullet item to check wrapping and indentation when the line extends beyond the first measure. - A final bullet item with `inline code`. > This block quote is temporary filler for checking quote spacing and contrast. ```python def score(x, model): return model.predict_proba(x)[:, 1] ``` ## BibTeX ```bibtex @online{aussetMethodesApprentissageStatistique2021, title = {Méthodes d'apprentissage Statistique Pour l'analyse Prédictive Du Risque de Crédit}, author = {Ausset, Guillaume}, date = {2021}, url = {http://www.theses.fr/s197391}, abstract = {L'objectif de la thèse est de développer des techniques d'apprentissage de modèles prédictifs du risque de crédit permettant d'exploiter de nouvelles sources de données afin de définir une approche Point In Time pour une estimation plus performante des risques afférents. Ces sources originales de données incluent par exemple les messages Bloomberg ou Twitter, dont le caractère éventuellement prédictif sera exploré à travers le projet. Des statistiques sur le volume de news Bloomberg citant une société existent déjà, mais ne proposent pas de traitement qualitatif du message (positif ou négatif). Aussi les méthodes développées s'adapteront à la structure particulière des données, i.e., caractère séquentiel et multitâche, et grande dimension de l'espace des features. Le traitement de nouvelles sources de données par des méthodes appropriées permettra de créer de nouveaux signaux Point-In-Time pour élaborer des stratégies de hedging ou des stratégies d'investissement dans le marché de crédit et ceci en complément des ratings.}, organization = {{http://www.theses.fr}} } ```