Predictive Data Quality Modelling for Financial Customer Data Using Machine Learning
Budget: $10 – $70 USD
I need an end-to-end ML experiment to predict duplicate customer records in a financial dataset from Kaggle. The goal is to build a proactive classification model that flags likely duplicates before they reach reporting, analytics, or risk pipelines.
The workflow should include data loading, EDA, synthetic duplicate labelling (since labels won’t exist), feature engineering, model training, and evaluation. Duplicate pairs will be created using techniques like exact duplication, small perturbations, and formatting inconsistencies.
Features should include exact matches, numeric differences (age, income, spending), and similarity measures. Models to test include Logistic Regression, Random Forest, Gradient Boosting, XGBoost, or similar, but deliver one final tuned model.
Evaluation should focus on F1-score (target ≥0.85), with a balance between precision and recall.
Deliverables: reproducible notebook, clean code, short report, and README.
The workflow should include data loading, EDA, synthetic duplicate labelling (since labels won’t exist), feature engineering, model training, and evaluation. Duplicate pairs will be created using techniques like exact duplication, small perturbations, and formatting inconsistencies.
Features should include exact matches, numeric differences (age, income, spending), and similarity measures. Models to test include Logistic Regression, Random Forest, Gradient Boosting, XGBoost, or similar, but deliver one final tuned model.
Evaluation should focus on F1-score (target ≥0.85), with a balance between precision and recall.
Deliverables: reproducible notebook, clean code, short report, and README.