01 / PROBLEM STATEMENT
Why this work exists
Manual review of accounting journals makes it difficult to prioritize potentially irregular entries. The project explores supervised classification as an auditor-assistance workflow.
Data and source material
- Multiple accounting-journal exports aligned to the Moroccan accounting plan
- Cleaned, encoded, and scaled derivative datasets
- Prediction CSVs and serialized model artifacts
02 / SOLUTION OVERVIEW
From input to usable output
Clean and encode data, compare five classifiers and export predictions for review.
- 01
Explore missing values and correlations
- 02
Clean, encode, and scale journal fields
- 03
Train model families in separate reproducible scripts
- 04
Export predictions for downstream review
03 / TECHNICAL ARCHITECTURE
Implemented components
Journals → preprocessing → classifier pipelines → predictions → report
04 / METHODOLOGY & IMPLEMENTATION
Decisions that shape the system
- The repository compares Decision Tree, Random Forest, Gradient Boosting, Logistic Regression, and K-Nearest Neighbours.
- A correlation matrix supports feature inspection.
- The system is an anomaly-triage prototype; predictions are not treated as confirmed fraud.
Implemented capabilities
- Preprocessing
- Five classifier families
- Saved artifacts
- Prediction exports
- Docker
05 / RESULTS & EVIDENCE
What the source supports
- Five classifier families and stored prediction artifacts are documented.
- The public README does not provide a validated common test table, so no accuracy or F1 figure is published here.
Results preserve the repository’s units, evaluation framing, and stated limitations. Missing evidence is not converted into a claim.
06 / CHALLENGES & LIMITATIONS
Where the evidence stops
- Public data files may reflect real accounting exports and are not reproduced in this portfolio.
- The repository does not document a leakage-safe shared evaluation protocol or complete common metrics.
- Model predictions require professional audit review.
07 / WHAT THIS DEMONSTRATES
Engineering signal
Accounting-data preprocessing, supervised model comparison, artifact export, and an appropriately assistive—not accusatory—use of machine learning.
08 / SOURCE & RELATED WORK
Trace the evidence
b289b1109dc8de9e47124ecd8eaa66228db0e358 ↗README.md · data_preprocessing.py · data_understanding.py · Report/