Abdulelah Abuabat

I work closely with supply, marketing, and partnerships at Mrsool, addressing data challenges through analysis and machine learning to offer insights and solutions for strategic decisions. I create analytical reports on various topics, including performance, following a workflow that includes defining needs, suggesting solutions, collecting data (mainly from our Redshift data warehouse via SQL), and delivering in-depth analyses and trends. This process leads to dynamic dashboards and predictive models integrated into our operations using platforms like AWS C2 for efficient deployment. Some of the projects I am participating in are:

• Fraud Detection: Developed AI models for detecting fraudulent transactions, enhancing the fraud team's ability to identify suspicious activities. Implemented both transaction-based and user-based (courier) models, leading to improved identification of abnormalities.
• Customer Churn Forecasting: Developed a hierarchical machine learning model to predict customer churn and define effective interventions for retention.
• Orders Forecasting (Demand Side): Developed a multivariate time-series forecasting model to predict location-based demand. Integrated the results with a heat map using a powerful open-source geospatial analysis tool, enhancing visualization.
• Freelancers-Couriers Churn (Supply Side): Developed a survival analysis model to predict the availability of freelance couriers, aiding in efficient demand management.
• Experiments Design (A/B Testing): Led the setup and analysis of A/B tests to assess the impact of layout changes and new incentive mechanisms.
• Causal Analysis for Marketing Activities: Led a quasi-causal analysis to evaluate the magnitude of global promotions on store profitability. Focused on understanding the impact of delivery discounts on overall profit increases, applying advanced statistical methods to derive actionable insights.
• Batching Orders Optimization: Built an algorithm for batching orders, optimizing service efficiency and courier profits during peak hours.
• Building a De-Normalized Schema for Reporting Purposes: I assisted in building a de-normalized schema to improve the efficiency and performance of data retrieval, which is crucial in reporting and data analytics.

At GaStat, the Big Data team focused on enhancing statistical indicators published quarterly, utilizing advanced statistical learning methods for forecasting and exploring non-traditional data sources for indicator verification. My contribution included developing a fraud detection model for the International Trade Index, addressing data integrity issues from customs data. The model, trained on historical data, considers multiple factors such as inflation and price variance by goods type to identify irregular entries, thereby improving the accuracy of published data.

I was working as a ML Engineer to develop ML and DL (deep learning) models for image classification-related issues. The biggest project was multi-source vision-based robotics, owned by the Bossa Nova company incorporated with several international giant retail companies, one of which was Walmart Inc. for their retail data solution robot. The main two projects were:

• Real-time AI for Robotic Inventory Analytics: Classifying Shelf and Price Tags with Advanced Deep Learning. Our project aimed to create a Robotic Inventory Analytics Engine (RIAE) to streamline inventory management. The task involved a range of activities, from item labeling to sophisticated object classification. I led the development of a shelf tag classifier, employing advanced neural networks such as ResNet and VGG, and utilized a high-performance Nvidia GTX 1080 GPU for efficient model training. Our efforts culminated in achieving an impressive F1 score of approximately 95%, reflecting the high accuracy and reliability of our system, which was successfully implemented in the market.
• Deep Detection and Classification Framework for Abnormal Tissues: The project aimed to optimize the classification of abnormal cervical cells (Pap smear) using deep convolutional neural networks (CNNs), leveraging small sample sizes for feature learning. To address dataset limitations, generative approaches like GANs were employed. The developed model is now utilized by physicians at the University of Pittsburgh Medical Center for automatic detection and classification of abnormal cervical cells.

Analyze and design information systems solutions for organizations to operate more efficiently. Tasks included analyzing and evaluating existing systems, communicating with stakeholders, and building the system specifications document for the suggested system. Some of the projects that I was working on were:

• Human Resource Management System for Small Companies: An advanced SaaS solution for enterprise planning and human management systems targeted at small and medium companies.
• Subsidization Eligibility: Building a subsidization eligibility engine. (External Project for the Ministry of Finance).
• Smart Police Car: designing a request for a proposal (RFP) for security companies to build a modified version of the police cars.
• Registration and Evaluation System for Car Accidents: Building a registration and evaluation system for car accidents for the traffic department.

• Classified Project.