Hello!
I'm MD. Usman Gani Joy, an enthusiastic researcher and a passionate developer researching on AI and ML.
Get in touch usmanjoycse@gmail.com
ORCID ID:https://orcid.org/0009-0003-9498-3828
Hello! I'm Usman Gani Joy, a machine learning enthusiast with an M.Sc. in Computer Science and Engineering from East Delta University, where I graduated with a CGPA of 3.83/4.00. My 1st paper was published in IEEE Access and currently I am researching and working on several papers as well. Click here to see my IEEE paper. I work as a Machine Learning Engineer and AI Consultant, specializing in developing predictive models and data-driven solutions.
My expertise includes big data, semi-supervised learning, transfer learning, deep learning, and customer churn analysis.
Academic Contributions: I am also an IEEE Access Reviewer, committed to advancing knowledge in the field. Click to go to my ORCID ID Link.
- Flutter
- React
- Redux
- Node.js
- Python
- Asp Net Core Web API
- Scaling
- Microservice
- Monolith
- Pytorch
- Tensorflow
- Keras
- Scikit-learn
- Pandas
- Numpy
- Pyspark
Gained foundational expertise in .NET Web API development, implementing core functionalities and enhancing backend efficiency through optimized API interactions.
Advised on AI projects, providing data-driven insights and developing predictive model applications for diverse industries.
Developed and optimized machine learning models for clients in diverse industries, leveraging cutting-edge algorithms and domain-specific techniques.
Proposed a big data-driven hybrid model combining deep neural network and machine-learning techniques to efficiently forecast customer churn in the streaming service industry. The model utilized LSTM, GRU, and Light GBM to capture sequential patterns and leverage insights for accurate churn prediction. Feature selection methods like Chi-squared testing and Sequential Feature Selection were employed, and the model interpretations were provided using SHAP and Explainable Boosting Machine (EBM). Extensive experiments demonstrated the hybrid model's superior performance with 95.60% AUC and 90.09% F1 score.
Employed deep neural network UNet for accurate CT liver tumor segmentation. Adapted VGG-16 for image classification to enhance UNet's performance.
Utilized XGBoost, Random Forest, and LSTM deep learning model to forecast wind power from real-world Patenga data. Co-authored a forthcoming paper on the project.
This paper explores innovative techniques for efficient model training with scarce labeled data, leveraging a semi-supervised approach.
Focuses on novel strategies for uncertainty quantification and promoting diversity in sample selection to improve semi-supervised learning frameworks.
Introduces a novel hybrid model integrating Neural ODEs with the Adams-Bashforth method to enhance forecasting performance in energy consumption prediction.