ADVANCED DATA SCIENTIST – FRAUD & FORENSICS
Level: Advanced / Senior
CapCircle is recruiting an experienced Data Scientist for a fintech fraud detection and model modernisation programme.
This role focuses on behavioural and historical interaction feature engineering, fraud scheme analysis, unsupervised anomaly detection, model selection, drift control and performance benchmarking. The successful candidate must have strong hands-on experience working with transactional data and fraud or financial crime use cases.
ROLE OBJECTIVE
Support the enhancement of an existing social engineering fraud model by developing stronger behavioural, transactional and historical interaction features. The objective is to improve precision and recall, reduce false positives and create an explainable, measurable and continuously monitored fraud detection capability.
KEY RESPONSIBILITIES
- Review and refine the existing fraud modus operandi and scheme library.
- Document known and emerging fraud schemes, relevant features, correlations and data gaps.
- Analyse large transactional datasets to identify suspicious behavioural patterns and fraud indicators.
- Engineer behavioural, transactional and historical interaction features for model training and scoring.
- Use prior transaction and communication patterns to distinguish established customer relationships from suspicious or newly created interactions.
- Develop, test and evaluate unsupervised anomaly-detection approaches on transactional data.
- Support supervised and unsupervised model selection, threshold calibration and champion-model promotion.
- Work with graph-derived features such as centrality, communities, multi-hop flows, fan-in, fan-out and repeat-offender linkages.
- Support entity linking and the development of explainable risk scores.
- Identify patterns associated with funnels, hubs, enticement loops, rapid movement, layering, collusion and suspicious off-ramping.
- Build point-in-time-correct feature datasets and prevent data leakage during training and evaluation.
- Ensure behavioural, historical and graph features can be used consistently across training, recurring scoring and inline decisioning.
- Conduct before-and-after model performance benchmarking using precision, recall, F1 score and false-positive rate.
- Analyse model errors and recommend improvements based on false positives, false negatives and investigator outcomes.
- Support drift monitoring, drift-triggered retraining and controlled model promotion.
- Implement deduplication and data-integrity controls across the fraud detection pipeline.
- Integrate fraud investigator labels and outcomes into model evaluation and retraining.
- Contribute to model documentation, feature definitions, versioning, audit lineage and end-to-end validation.
- Work closely with the Senior Graph Data Scientist / ML Lead, MLOps Engineer, Solution / Data Architect and fraud stakeholders.
ESSENTIAL EXPERIENCE AND TECHNICAL REQUIREMENTS
- Proven data science experience within fraud detection, financial crime, banking, payments or fintech.
- Hands-on experience with unsupervised anomaly detection using transactional data.
- Advanced Python skills for data analysis, feature engineering and machine learning.
- Strong experience with large, complex transactional datasets.
- Experience developing behavioural and historical interaction features.
- Strong knowledge of fintech fraud typologies, social engineering fraud and financial crime patterns.
- Experience with model selection, model validation, threshold optimisation and performance benchmarking.
- Strong understanding of precision, recall, F1 score, false-positive rate and class-imbalance challenges.
- Experience with model monitoring, data drift, concept drift and retraining controls.
- Understanding of point-in-time correctness, data leakage, train-and-serve consistency and feature versioning.
- Knowledge of MLOps best practices and the requirements for moving models into production.
- Strong analytical, problem-solving and documentation skills.
- Ability to explain model results, risk indicators and performance trade-offs to technical and non-technical stakeholders.
ADVANTAGEOUS EXPERIENCE
- Experience with NetworkX, Neo4j or graph-derived fraud features.
- Experience with PostgreSQL, feature stores, Redis, Kafka, Docker, Airflow or equivalent technologies.
- Experience integrating investigator feedback into model improvement and retraining.
- Exposure to real-time or near-real-time fraud decisioning.
WAYS OF WORKING
- Work within an Agile technology team focused on continuous delivery.
- Participate in sprint planning, daily stand-ups, sprint reviews, retrospectives and demonstrations.
- Take ownership of analysis, experiments, feature definitions and assigned deliverables.
- Collaborate across data science, engineering, architecture, fraud and investigation teams.
- Communicate findings, risks, assumptions and recommendations clearly.
WHO SHOULD APPLY?
- This position is suited to an advanced Data Scientist who has worked directly with fraud or financial crime data and can demonstrate hands-on anomaly detection and feature-engineering experience. General reporting, business intelligence or entry-level data science experience will not meet the requirements.
- Apply through CapCircle. Your CV must clearly show your fraud or financial crime experience, anomaly-detection work, Python and modelling skills, datasets used, performance improvements achieved and personal contribution to production solutions.
ADDITIONAL MANDATORY REQUIREMENTS
- A minimum of 5–8 years’ relevant professional experience is required.
- Candidates must hold a relevant tertiary degree and applicable professional or technical certifications related to the position.
- These are senior and advanced-level positions and are not suitable for interns, recent graduates or students.
- Candidates must be able to work on a hybrid basis from the office in Roodepoort, Gauteng, South Africa.
- Office attendance is required three days per week and is non-negotiable.
- Please only apply if you meet the required experience, qualification and hybrid working requirements.