🤖
Machine learning models
From cohort building and feature engineering to benchmarking, tuning and calibrating models on large, messy, real-world data.
Open to Data Science & ML roles
Hello, I'm Shadab Ansari
Machine Learning Engineer
I build explainable, reproducible machine learning systems, from raw data to a deployed app. M.Tech (CS & IT) at MANUU, Hyderabad, researching cancer survival prediction on large clinical registry data.
🎓 M.Tech @ MANUU
🔍 Explainable AI
⚙️ MLOps
Pythonscikit-learnXGBoostLightGBMSHAPLIMElifelinesscikit-survivalPandasNumPySQLMLflowDVCDockerKubernetesFlaskGitHub ActionsPlotly
Turning data into decisions people can trust
I'm an M.Tech (Computer Science & IT) student at Maulana Azad National Urdu University (MANUU), Hyderabad. My research applies machine learning to colorectal cancer survival prediction under the guidance of Dr. Mohd Omar.
I care about models that are not only accurate but also explainable (so a clinician can see why a prediction was made), validated on data they have never seen, and reproducible end to end: versioned data, tracked experiments, tests, CI and containerised deployment.
Before moving into ML I built frontend projects with HTML, CSS and JavaScript, so I'm comfortable shipping the last mile too: turning a model into a web app that people can actually use.
☕ Coffee enthusiast · 🌳 Nature lover · 🎵 Music keeps me focused
🔍
Explainable AI
SHAP, LIME and partial dependence to show why a model made each prediction, so people can check it and trust it.
🚀
MLOps & deployment
DVC pipelines, MLflow tracking, CI with GitHub Actions, and Flask apps in Docker and Kubernetes.
My tech stack
Tools I've used in real projects, grouped by where they fit in the ML lifecycle.
🐍
Languages & Data
- Python
- SQL
- MySQL
- NumPy
- Pandas
- Jupyter
- EDA
🤖
Machine Learning
- scikit-learn
- XGBoost
- LightGBM
- Gradient Boosting
- Random Forest
- SVM
- Optuna
- Cross-validation
- Calibration
- Imbalanced data
🔍
Explainable AI
- SHAP
- LIME
- Partial dependence
- Permutation importance
📈
Survival Analysis & Stats
- lifelines
- scikit-survival
- Cox PH
- Random Survival Forest
- Kaplan-Meier
- C-index
- Bootstrap CIs
- Decision curves
- SciPy
⚙️
MLOps & Deployment
- MLflow
- DVC
- DagsHub
- pytest
- GitHub Actions
- Docker
- Kubernetes
- Flask
- Gunicorn
- REST API
🎨
Visualization, Web & Tools
- Matplotlib
- Seaborn
- Plotly
- HTML5
- CSS3
- JavaScript
- React
- Git
- GitHub
- VS Code
Featured work
★ M.Tech Research
View code
Explainable ML for 5-Year Colorectal Cancer Survival Prediction
An end-to-end, reproducible ML pipeline that predicts five-year overall survival for colorectal cancer patients and explains each prediction. Trained on the SEER 17 cancer registry and externally validated on AACR Project GENIE.
257,827patients, SEER 17 registries
0.877test ROC-AUC (95% CI 0.874–0.880) on 50,338 held-out patients
0.878temporal validation AUC (train 2010–15, test 2016–18)
0.790external validation AUC, AACR Project GENIE
0.809best C-index across 5 survival models
+0.042AUC over the stage + age + sex clinical baseline (0.835)
What I did
- Built the cohort from 324,174 registry records (cleaning, exclusions, feature engineering) down to 257,827 patients.
- Benchmarked 6 classifiers (Logistic Regression, SVM, Random Forest, Gradient Boosting, XGBoost, LightGBM) with stratified 5-fold CV and 1,000-sample bootstrap confidence intervals.
- Checked calibration (slope 1.02), decision curves and subgroup performance, not just AUC.
- Compared 5 survival models: Cox PH, CoxNet, Random Survival Forest, Gradient Boosting Survival, XGBoost-Cox.
- Explained predictions globally and per patient with SHAP, LIME and partial dependence plots.
How it ships
- DVCprepare → train → evaluate pipeline with params.yaml
- MLflowexperiment tracking and Model Registry on DagsHub
- CIGitHub Actions: pytest on synthetic data, no-patient-data guard, Docker build
- DockerFlask + Gunicorn app with /predict and /health, Kubernetes manifests
- Appcalibrated 5-year survival probability + top SHAP factors per patient
Figures from the study click to enlarge
- Python
- scikit-learn
- XGBoost
- LightGBM
- SHAP
- lifelines
- scikit-survival
- MLflow
- DVC
- Flask
- Docker
- Kubernetes
- GitHub Actions
More projects
Research & writing
Explainable, Calibrated and Externally Validated Machine Learning for Five-Year Colorectal Cancer Survival
Prepared for submission to ICCIDSDT-2027 (Springer LNNS). An extended journal version with GENIE external validation is in preparation.
Writing on Medium
I write about machine learning, model explainability and what I learn along the way.
Read on MediumEducation & achievements
-
Pursuing · 2nd year
M.Tech, Computer Science & Information Technology
Maulana Azad National Urdu University (MANUU), Hyderabad
Research: explainable machine learning for colorectal cancer survival prediction. Guide: Dr. Mohd Omar.
-
2022 – 2024
Master of Computer Applications (MCA)
Maulana Azad National Urdu University (MANUU), Hyderabad
CGPA 8.4 / 10
3rd Rank
Coding Ninjas Coding Byte Contest
Coordinator of Cultural Activities
MANUU
Member, Food-cum-Audit Committee
Boys' Hostel, MANUU (2023–24)
7-Day Bootcamp Certificate
Dev Town with Microsoft Learn and GDSC KIIT
Let's build something together
I'm looking for Data Scientist, Machine Learning Engineer and AI/ML research roles. My inbox is always open.
Emailshadabahamed63@gmail.com
LinkedInin/shadab-ansari-dln
GitHubgithub.com/shadabdln
LocationHyderabad, Telangana, India