ShockLab Seminar Series - Akamba Mani Crescence Catherine

The next Shocklab Seminar will be held online on Wednesday, 10 December 2025, from 16:00 - 17:00. Akamba Mani Crescence Catherine will be presenting Credit Risk Prediction in Peer-to-Peer Lending Platforms using Graph Features. Please come along if you can, it’s sure to be an engaging talk!

 

 

Credit granting plays a crucial role in economic development by enabling enterprise growth or by making individual life easier. Credit granting is guaranteed by traditional institutions or lending platforms. However, due to the lack of relevant information, they are often weakened by high default rates, which has led to the suspension of their activities. To solve this problem, we propose a complete protocol for predicting credit risk by integrating new descriptors extracted from the bipartite loan-modality graph and the modality-modality graph, both constructed from categorical attributes and numeric attributes made categorical by supervised and unsupervised discretization processes. The personalized PageRank is applied to these graphs for each loan, and the values obtained are incorporated as new input descriptors for classic credit risk prediction models. Experiments were carried out on two datasets, considering six standard credit risk prediction models and two metrics for assessing model performance. Shapley values are considered to assess the importance of the new descriptors. The results show that the combinations we propose deliver improvements ranging from 0.05% to 3.49% for the accuracy metric and from 0.01% to 37.3% for the F1-score metric. In addition, the new proposed descriptors occupy 100% and 80% of the Top-10 best Shapley values in the two datasets considered.

Bio:

My name is AKAMBA MANI Crescence Catherine, and I hold a Master's degree in Computer Science, specializing in Data Science, from the University of Yaoundé 1 in Cameroon. My thesis focused on predicting credit risk in peer-to-peer lending platforms using attributes extracted from graphs. In this project, we proposed constructing the bipartite loan-modality graph and the modality graph from the descriptive attributes of borrowers, and applying a customized PageRank to these graphs to extract new descriptors. The performance analysis of the prediction models confirmed that the new descriptors not only improve predictions but are also among the most important according to Shapley value analysis. The bulk of this work will be submitted to the journal Advanced Engineering Informatics. My field of research has also expanded to include several projects, notably the improvement of facial recognition for online authentication using the Deep Retinex-Net model (which allows for the processing of low-light images), the FDN (Feature Decoupling Network) model for facial movement analysis, and the CNN (Convolutional Neural Network) model for emotion detection. I also worked on a recommendation system for public transportation in the city of Yaoundé, Cameroon. We started by collecting comprehensive data on the city's population, then developed a multiple logistic regression model to predict transportation fares for a given route. I have also worked in the field of parallel machine learning.