Soufian Boukir
Soufian BoukirJunior ML & Data science Specialist

About me

Passionate about applying machine learning, data analysis, and statistical modeling to solve real-world problems.

My interests span the entire machine learning lifecycle, from data engineering and model development to deployment and MLOps. By combining software engineering with AI, I build scalable, data-driven solutions that deliver practical value.

Currently I'm building my career in Data Science and Applied AI, with a strong focus on production-ready intelligent systems.

Currently Focusing On

  • Advanced Machine Learning: Deepening my understanding of ML, model optimization, feature engineering, ensemble methods, and modern deep learning techniques.

  • MLOps & Machine Learning in Production: Learning how to build, deploy, monitor, and maintain scalable machine learning systems using industry best practices, automation workflows, model versioning, and cloud-based infrastructure.

  • Applied Maths: Strengthening my foundations in linear algebra, calculus, probability, statistics, and optimization to better understand machine learning algorithms and improve analytical problem-solving skills.

Projects Done

Food Delivery Intelligence System with Predictive Analytics

Abstract:
In this project, we developed an end-to-end data science pipeline for a food delivery dataset containing over 15,000 records. The workflow includes data cleaning, exploratory data analysis (EDA), and predictive modeling. Regression models were used to predict delivery time, while classification models were applied to determine whether an order is likely to be canceled. In addition, an interactive Streamlit dashboard was built to visualize insights and allow real-time interaction with the predictive models.

Keywords:
Food Delivery Analytics, Machine Learning, Business Intelligence

Links:
View Project Report • GitHub • Live Dashboard

Data-Driven E-commerce Intelligence: Customer Analytics and Revenue Forecasting System

Abstract:
This project presents an end-to-end e-commerce analytics and forecasting platform built using a synthetic Amazon-style dataset of 100,000 transactions (2020–2024). The system integrates sales analysis, customer segmentation, product analytics, and revenue forecasting within an interactive Streamlit dashboard. Among four evaluated models, XGBoost achieved the best performance (RMSE: 61,570, MAPE: 3.71%). The forecasts highlight stable but stagnant revenue trends through 2026, suggesting the need for strategies such as market expansion, discount optimization, and customer retention programs.

Keywords:
E-commerce Analytics, Time Series Forecasting, Business Intelligence

Links:
View Project Report • GitHub • Live Dashboard

Robustness Analysis of Deep Learning Models Under Controlled Image Corruptions

Abstract:
This study evaluates the robustness of deep learning models (MLPs, CNNs, and ResNets) under different image corruption conditions. The models were tested on CIFAR-10 using various real-world corruptions such as noise, blur, and compression at different severity levels. Performance was measured using accuracy, precision, recall, F1-score, and robustness-specific metrics such as accuracy drop and sensitivity. Results show that while deeper models perform better on clean data, they are less stable under corrupted inputs compared to simpler architectures.

Keywords:
Robustness, Image Corruption, CIFAR-10, Deep Learning, CNN

Links:
View Project Report • GitHub