Foresight & Market Positioning
Predicting Credit Default Risk
Building a high-accuracy model to predict credit defaults, helping American Express balance risk management with a better customer experience.
Year :
2023
Industry :
Financial Services, Data Science & Risk Analytics
Client :
Self: Amex Competition
Project Duration :
3 weeks
The Challenge:
Credit default prediction is at the heart of managing risk in consumer lending. For American Express, improving these models means not only reducing financial losses but also creating a smoother, fairer customer experience.
In 2022, American Express launched a global competition to build a machine learning model that could outperform their existing production model, using industrial-scale, time-series behavioural data and anonymised customer profiles.
Insight:
A customer’s likelihood to default is influenced by a complex interplay of spending, payment, balance, delinquency, and risk behaviours over time. By analysing these patterns holistically, it’s possible to make more accurate predictions, enabling smarter lending decisions and more personalised customer experiences.
Approach:
Data Exploration & Feature Engineering: Analysed anonymised variables across delinquency (D_), spend (S_), payment (P_), balance (B_), and risk (R_*) categories.
Time-Series Modelling: Incorporated sequential behavioural data to capture trends and shifts in customer activity.
Model Development & Validation: Built and tuned a predictive model with evaluation metrics between 0.70–0.95, ensuring no overfitting or underfitting.
Testing on Unseen Data: Achieved 83% accuracy in predicting defaults, demonstrating strong generalisation capability.
Summary :
The model provided a more precise tool for identifying potential defaults, enabling:
Better Risk Management – Reducing exposure while maintaining healthy approval rates.
Improved Customer Experience – Allowing more customers to be approved without increasing risk.
Business Efficiency – Supporting data-driven credit policy decisions.
By delivering a model with strong predictive power, this project showcased how advanced analytics can directly improve both business economics and customer satisfaction in financial services.
More Projects
Foresight & Market Positioning
Predicting Credit Default Risk
Building a high-accuracy model to predict credit defaults, helping American Express balance risk management with a better customer experience.
Year :
2023
Industry :
Financial Services, Data Science & Risk Analytics
Client :
Self: Amex Competition
Project Duration :
3 weeks
The Challenge:
Credit default prediction is at the heart of managing risk in consumer lending. For American Express, improving these models means not only reducing financial losses but also creating a smoother, fairer customer experience.
In 2022, American Express launched a global competition to build a machine learning model that could outperform their existing production model, using industrial-scale, time-series behavioural data and anonymised customer profiles.
Insight:
A customer’s likelihood to default is influenced by a complex interplay of spending, payment, balance, delinquency, and risk behaviours over time. By analysing these patterns holistically, it’s possible to make more accurate predictions, enabling smarter lending decisions and more personalised customer experiences.
Approach:
Data Exploration & Feature Engineering: Analysed anonymised variables across delinquency (D_), spend (S_), payment (P_), balance (B_), and risk (R_*) categories.
Time-Series Modelling: Incorporated sequential behavioural data to capture trends and shifts in customer activity.
Model Development & Validation: Built and tuned a predictive model with evaluation metrics between 0.70–0.95, ensuring no overfitting or underfitting.
Testing on Unseen Data: Achieved 83% accuracy in predicting defaults, demonstrating strong generalisation capability.
Summary :
The model provided a more precise tool for identifying potential defaults, enabling:
Better Risk Management – Reducing exposure while maintaining healthy approval rates.
Improved Customer Experience – Allowing more customers to be approved without increasing risk.
Business Efficiency – Supporting data-driven credit policy decisions.
By delivering a model with strong predictive power, this project showcased how advanced analytics can directly improve both business economics and customer satisfaction in financial services.
More Projects
Foresight & Market Positioning
Predicting Credit Default Risk
Building a high-accuracy model to predict credit defaults, helping American Express balance risk management with a better customer experience.
Year :
2023
Industry :
Financial Services, Data Science & Risk Analytics
Client :
Self: Amex Competition
Project Duration :
3 weeks
The Challenge:
Credit default prediction is at the heart of managing risk in consumer lending. For American Express, improving these models means not only reducing financial losses but also creating a smoother, fairer customer experience.
In 2022, American Express launched a global competition to build a machine learning model that could outperform their existing production model, using industrial-scale, time-series behavioural data and anonymised customer profiles.
Insight:
A customer’s likelihood to default is influenced by a complex interplay of spending, payment, balance, delinquency, and risk behaviours over time. By analysing these patterns holistically, it’s possible to make more accurate predictions, enabling smarter lending decisions and more personalised customer experiences.
Approach:
Data Exploration & Feature Engineering: Analysed anonymised variables across delinquency (D_), spend (S_), payment (P_), balance (B_), and risk (R_*) categories.
Time-Series Modelling: Incorporated sequential behavioural data to capture trends and shifts in customer activity.
Model Development & Validation: Built and tuned a predictive model with evaluation metrics between 0.70–0.95, ensuring no overfitting or underfitting.
Testing on Unseen Data: Achieved 83% accuracy in predicting defaults, demonstrating strong generalisation capability.
Summary :
The model provided a more precise tool for identifying potential defaults, enabling:
Better Risk Management – Reducing exposure while maintaining healthy approval rates.
Improved Customer Experience – Allowing more customers to be approved without increasing risk.
Business Efficiency – Supporting data-driven credit policy decisions.
By delivering a model with strong predictive power, this project showcased how advanced analytics can directly improve both business economics and customer satisfaction in financial services.