Certification projects — Concepteur Développeur en Science des Données (RNCP35288, Level 6) Jedha Bootcamp · Obtained May 2026

This organisation gathers all projects developed and validated during the CDSD certification programme. Each project covers a complete data science pipeline — from data collection and exploratory analysis to machine learning modelling, deployment and documentation.

Crime trend analysis and forecasting in France

End-to-end ML pipeline on open government data (data.gouv.fr): feature engineering with cyclic encoding, lags and rolling averages, benchmark of 5 models, LightGBM champion (R² 0.979 on test set, 0.978 ± 0.002 in temporal cross-validation without data leakage). MLflow tracking, Docker multi-stage, FastAPI REST API and Streamlit dashboard deployed on Hugging Face Spaces. Ethical and GDPR framing documented.

Python LightGBM XGBoost Prophet MLflow FastAPI Docker Streamlit Plotly

🔗 Live demo

Rental pricing optimisation and check-in delay analysis

Analysis of the impact of check-in delays on revenue and customer satisfaction. Price prediction model with MLflow pipeline, FastAPI REST endpoint and Streamlit interface for real-time fare estimation.

Python Pandas Scikit-learn Random Forest MLflow FastAPI Streamlit Plotly Express

Travel destination recommendation engine — Cloud ETL pipeline

Multi-source ETL pipeline on 34 French cities: Open-Meteo API for weather forecasts, Overpass API and web scraping for hotel data, AWS S3 data lake, PostgreSQL data warehouse on AWS RDS Aurora, transactional writes via SQLAlchemy. Interactive geospatial visualisation with Plotly Mapbox.

Python Pandas boto3 AWS S3 AWS RDS Aurora SQLAlchemy BeautifulSoup Plotly Mapbox

SMS classification by transfer learning

Rigorous comparison of three approaches: baseline classifier, TF-IDF with logistic regression, fine-tuned DistilBERT. Champion model: 99.5% accuracy, F1-spam 98.2%, 100% precision with zero false positives on 836 test messages. Documented error analysis on the 4 false negatives.

Python PyTorch Hugging Face Transformers DistilBERT Scikit-learn TF-IDF

Product catalogue clustering and e-commerce recommendation engine

NLP pipeline for product catalogue restructuring: SpaCy lemmatisation, TF-IDF vectorisation (unigrams and bigrams), cosine DBSCAN clustering, topic modelling via LSA. Recommender system suggesting similar products based on semantic similarity.

Python SpaCy TF-IDF DBSCAN TruncatedSVD Scikit-learn Pandas

Geospatial segmentation of 1.2M rides

Driver positioning optimisation using unsupervised learning on 1.2 million New York City rides. KMeans benchmark (elbow and silhouette methods) and DBSCAN with Haversine metric. Positioning recommendation table across 84 time slots.

Python Scikit-learn KMeans DBSCAN Pandas Plotly Matplotlib Seaborn

Weekly sales forecasting for 45 stores

Regularised regression pipeline with GridSearchCV and ColumnTransformer. Ridge and Lasso benchmarked against linear regression baseline. R² of 0.963 on test set with explicit overfitting control. Production artefacts exported and prediction endpoint operational.

Python Scikit-learn Ridge Lasso GridSearchCV ColumnTransformer Pandas Matplotlib

Inferential analysis of mutual attraction factors

Exploratory and inferential analysis of a speed dating dataset. Statistical tests: Welch, chi², Mann-Whitney. Interpretable logistic regression under Statsmodels to identify predictive factors of compatibility. Documented hypotheses and effect sizes.

Python Pandas NumPy Statsmodels Scikit-learn Scipy Matplotlib Seaborn

Real estate price prediction based on climate risk

Predictive modelling of real estate prices with a focus on the impact of climatic events. Identification of safe and profitable locations by combining property market data with environmental risk indicators.

Python Pandas Scikit-learn Plotly Matplotlib

All projects were developed, validated and defended as part of the Concepteur Développeur en Science des Données certification (RNCP35288, Level 6 — equivalent to Bachelor's degree), delivered by Jedha Bootcamp.

👤 Frédéric Tellier · LinkedIn