Wenfei Liu

Coursework: Deep Learning (Dist.), Machine Learning & Data Mining (Dist.), Applied Statistics, AI for Industry, Ethical AI Implementation

Research: 3 journal publications (1 first-author); 2 granted utility patents (CN 214272475 U; CN 210658826 U)

• Evaluated six models (LR, FFN, Isolation Forest, Autoencoder, Ensemble) on a 284k-transaction dataset under 1:578 class imbalance; ENS achieved PR-AUC 0.772 under strict chronological split
• Applied SHAP, PSI drift analysis across 29 features, and isotonic calibration; produced compliance-aware report on GDPR / EU AI Act implications

• Fine-tuned DistilBERT to Macro F1 0.931 and ROC-AUC 0.989 (+4.5% F1 over TF-IDF baseline); structured around business threshold recommendations for non-technical stakeholders

• Selected Logistic Regression over Random Forest on cost-curve robustness (ROC-AUC 0.818, t = 0.48); applied RFECV, SMOTE, threshold optimisation, and post-hoc isotonic calibration across 253,680 records
• K-means clustering (k=3, PCA-6) identified three clinically differentiated health profiles; SHAP confirmed GenHlth, Age, BMI as dominant risk drivers

• Managed end-to-end delivery for commercial fit-out projects; coordinated contractors, vendors, and stakeholders from client brief through to completion

• Mapped satellite imagery for MSF in Delta Amacuro, Venezuela, tracing buildings and road networks to support disaster response planning