Frédéric Ferreira

Collection, cleaning, and analysis of your data to extract actionable insights that will optimize your decisions.

Design, training, evaluation, deployment, and monitoring of machine learning algorithms to solve your business problems.

Completion of professional projects validating the Data Scientist title, including:

Formalizing user stories, analyzing project-inherent risks, evaluating human, financial, and technical resource needs, sprint planning.

Renting EC2 virtual machines, storing data in S3, managing access with IAM, creating EMR clusters in the European territory (GDPR compliance), connecting via SSH to the master machine, executing processing with PySpark.

Managing the ML model lifecycle with MLflow (designing a training pipeline, evaluating and storing models).

Continuous design and deployment on the cloud (github actions + hosting) of a Django API that queries the best-performing model. Monitoring the model in production via Data Drift analysis.

Creation and deployment of an application (Streamlit, Dash) allowing a bank customer relationship manager to visualize customer information and call the credit scoring API to approve or reject a loan application.

Feasibility study of an article classification engine based on their description and/or image.

Use of basic NLP techniques (Bag of words, Tf-Idf) and advanced techniques (Word2vec, BERT, USE) for text feature extraction.

SIFT then CNN for image feature extraction.

Clustering of text/image features then comparison with actual classes (ARI score)

Supervised classification of articles using a pre-trained CNN.

RFM and customer satisfaction-based feature engineering

Clustering with Hierarchical Agglomerative Clustering, K-means, and DBSCAN

Analysis of clustering and its stability evolution.

Completion of 9 data analysis projects inspired by real-world work scenarios.

Collection, cleaning, and analysis of data on a Jupyter notebook.

Use of SQL and the SQLiteStudio and PostgreSQL RDBMS.

The study focused on global malnutrition and was conducted on a Jupyter notebook.

Reconciliation of exports, analysis of store sales + outlier detection.

Analysis of key figures, KPIs, visualizations, targeted customer analysis. Bivariate analysis of the most relevant variables.

Automation of a diagnostic report on gender equality through the creation of a KNIME workflow. Data anonymization to comply with GDPR.

Creation of an interactive dashboard with Power BI + Dax.

Exploratory data analysis + proposal of country groupings to target for international expansion. Clustering performed with HAC and K-means algorithms, followed by PCA to understand the groups and the links between countries.

Design and training of a supervised learning model for automatic classification of counterfeit banknotes using their characteristics (length, height, margin dimensions). Algorithm used: logistic regression.