How aayush-0131/12-jha-amritanshu-reddy-codes works
Project type and purpose
• This is an academic/research‐oriented data‐science project, not a full enterprise product. Its goal is to detect credit‐card fraud by formulating a convex optimization problem that minimizes the combined cost of false positives and false negatives.
• Key indicators: README title “Credit Card Fraud Detection and Cost Optimization Using Statistical Optimization Techniques,” no source code folders (only report, slides, README).
(minimal) file structure
• Top level contains documentation artifacts (PDF report, Quarto file, PowerPoint, README), no code folder.
• Implies the “system” is conceptual: data ingestion → model formulation → solver → evaluation/reporting.
README for additional insights
• Confirms use of Linear Programming (LP) and Quadratic Programming (QP).
• Team focused on modeling and cost formulation rather than deploying a microservice or web front end.
a. Main components
1. Data Source: Historical credit‐card transaction dataset (could be CSV or database).
2. Data Ingestion & Preprocessing: cleaning, feature extraction.
3. Cost Function Module: defines cost of FP and FN.
4. Optimization Model Builder: formulates LP/QP based on features and cost.
5. Solver Engine: uses an LP/QP solver library (e.g., CVX, Gurobi, CPLEX).
6. Results & Evaluation: computes confusion matrix, total cost.
7. Reporting/Visualization: generates tables, plots, and the project report.
b. Relationships & interactions
• Data flows from Data Source → Preprocessing → Model Builder.
• Model Builder interacts with Cost Function Module to parameterize the optimization.
• Model Builder sends the problem spec to Solver Engine.
• Solver Engine returns predicted labels → goes to Results & Evaluation.
• Evaluation outputs metrics to Reporting/Visualization.
c. Architectural patterns/principles
• Modular separation: data, cost definition, modeling, solving, evaluation.
• Pipeline pattern (linear flow of data through stages).
• Clear separation of concerns supports swapping solvers or cost functions.
d. Technologies/frameworks (if this were implemented)
• Data processing: Python pandas or R data.frame.
• Optimization modeling: cvxpy, PuLP, or JuMP.
• Solver back‐ends: Gurobi, CPLEX, OSQP.
• Reporting: Quarto (QMD), LaTeX/PDF, matplotlib or ggplot2.
Final recommendations
• Be as granular as it makes sense—e.g., split Preprocessing into Cleaning vs. Feature Engineering if your report covers both.