Mohamed A

Data Analytics intern

Banks • Full-time

Jun 2025 - Aug 2025 • 2 mos

• Transaction data quality pipeline: built an end-to-end pipeline in Polars on a multi-million-row synthetic banking transactions dataset; converted CSV to Parquet for faster I/O, handled null values per column, removed duplicates, and clipped outliers at the 1st and 99th percentiles instead of dropping rows to preserve data volume. • Anomaly feature engineering: designed a currency-discrepancy feature comparing local vs. foreign amounts against the recorded exchange rate to flag transactions whose totals were inconsistent — a useful signal for downstream fraud or data-entry investigation. • Loan approval prediction (supervised ML): trained a Logistic Regression classifier in scikit-learn to predict loan approval; engineered a debt-to-income ratio feature, performed correlation analysis using Chi-square, Cramér's V, and point-biserial tests, applied one-hot encoding, and evaluated the model with accuracy, confusion matrix, precision, recall, and F1-score. • Customer segmentation (unsupervised ML): clustered marketing-campaign customers using KMeans with PCA for dimensionality reduction; selected k via the elbow method and interpreted cluster centroids to describe distinct customer profiles. • Fintech EDA: cleaned a lending dataset (missing-value imputation by group, outlier handling, datetime parsing), engineered installment and grade-binning features, and applied one-hot encoding for low-cardinality categoricals and label encoding for high-cardinality ones (US states). • Data wrangling fundamentals: completed practical labs on real-world datasets (Netflix titles, San Francisco building permits) using pandas, matplotlib, and scikit-learn's IterativeImputer for multivariate imputation.