Salaheldin Mohamed Gaffar

Risk analyst | Data ScientistFreelance / Project

• Job Details:• Developed statistical risk regression (GLM and Random Forest) models using R programming language for flood-human displacement to improve UNDRR’s disaster prediction analytics using public global data sets. • Trained, validated, and performed hyperparameter/ performance tuning on the statistical risk models, achieving 85% accuracy in matching geospatial analysis of 913 historical flood events spanning 163 countries. • Performed data collection (information retrieval using APIs), data mining and cleaning and applied feature engineering techniques to select 15 features out of 124 possible features from the World Bank country-specific vulnerability indicators for a given flood event. • Spearheaded the application of geospatial modeling to predict flood patterns and inform resource deployment during 3 ongoing disaster use cases; developed interactive dashboards with Power BI reports for clearly communicating results and modeling trade-offs to stakeholders (international partners).

Senior Machine Learning EngineerFull Time

• Job Details:• Led a development team of 4 engineers that created and validated an innovative C# - Python simulation workflow software product to predict lithium-ion batteries’ performance degradation with 95% accuracy, using statistical techniques for modeling. • Assisted product development team for the battery performance predicting software product, in coordination with our product owner and client. . • Incorporated python simulation workflows in battery pack design process of US automotive business partners, improving battery life by 15% and reducing vehicle total cost of ownership by 12%. • Built LSTM recurrent neural network (RNN) AI model using TensorFlow (Keras) for time-series forecasting of vehicle component temperatures with a maximum error of 2 degrees in temperature. • Trained the LSTM using Azure ML studio and deployed the LSTM deep learning (DL) forecasting model as part of the vehicle thermal protection design pipeline, which resulted in a 3% reduction in vehicle weight. • Implemented a k-means clustering model (using scikit-learn) to segment the species in a combustion simulation framework, cutting down simulation time by 35%.