Oct 2024 – Jul 2025 · 9 mos · Abu Dhabi, Abu Dhabi Emirate, United Arab Emirates · On-site

Dec 2023 – Sep 2024 · 9 mos · Abu Dhabi Emirate, United Arab Emirates · On-site

Nov 2019 – Nov 2023 · 4 yrs

• Customer targeting with Machine Learning: Predict list of clients that are more likely to get a loan in the next months using machine learning models (Random Forest, Logistic Regression, XGBoost, LDA…) thus reducing costs by making more efficient campaigns with higher conversion rates. This is achieved by learning from a history of characteristics and behavior of clients.
• Using GitLab for versioning of the project, source code management and monitoring.
• Parallelization of data manipulation tasks using Dask.
• Testing coded features with nose and unittest.
• Deploying the application using a Docker container.
• Automating the process of clients scoring, and providing the final ‘top’ list for business exploitation.

Feb 2019 – May 2019 · 3 mos · Province de Khouribga, Morocco

• Developing Deep Learning models using Keras and Tensorflow to predict when certain critical machines will fail, so that maintenance actions can be planned in advance, thus reducing maintenance and operational costs and improving the safety of employees. This is achieved by training a LSTM (Long Short Term Memory) model on a history of sensors (temperature, vibrations, pressure, etc..) and failures of each machine.
• 1- Pre-processing of sensors data using pandas and numpy:
• Deletion of outliers and empty values imputation.
• Aggregating data (due to the lack of computational power).
• Applying Principal Component Analysis to reduce the number of variables.
• Format raw data to 3D sequences for usage by LSTM networks.
• 2- Modeling using Keras, Tensorflow and HyperOpt:
• Using LSTM neural networks, a special kind of recurrent neural networks that have the ability to learn long-term dependencies using a cell state for information to flow through time unchanged, thus avoiding the vanishing gradient problem that the standard version suffers from.
• Classification Problem: predict a binary variable, whether or not the machine will fail in the next N days.
• Regression Problem: predict the amount of time remaining until the next failure.
• Hyper-parameter tuning of the models by using Bayesian optimization (a better and more efficient approach to finding the best set of hyper-parameters of the model than grid search or random search).
• 3- Developing a user interface for real time supervision of machines’ states using Dash.