Recent Projects
The Fall Prediction project leverages machine learning to analyze vital signs and sensor data from elderly users, aiming to predict and prevent fall incidents proactively. By applying advanced techniques such as feature engineering, cross-validation, and hyperparameter tuning, the project achieved a remarkable 90% reduction in fall incidents. This work not only improved model accuracy by 25% but also significantly enhanced safety and quality of life for individuals at Choice Support.
The Prediction of Electricity Billed Amount project harnesses machine learning to forecast billed amounts, significantly improving load prediction accuracy by 30% and reducing operational costs by 25%. Collaborating with engineering teams, predictive maintenance models were developed to prevent transformer overloading, extending asset lifespan by 15%. By processing over 1TB of substation data daily using SQL, Hadoop, and Cassandra, this project enabled data-driven decisions, enhancing overall system efficiency by 20%.
The Customer Churn Prediction project focused on identifying at-risk customers using machine learning, enabling personalized retention strategies that increased customer retention by 50%. Predictive models for demand forecasting reduced stock-outs by 30% and overstock by 25%, optimizing inventory management. Advanced customer segmentation and persona analysis enhanced marketing strategies, further improving retention rates. Additionally, time-series analysis of unit price patterns identified trends and anomalies, aiding in price prediction and refining sales strategies.
This project focuses on developing a machine learning solution to classify paragraphs of text into one of five specific topics: artificial intelligence, movies about artificial intelligence, programming, philosophy, or biographies.
The objective is to build a model that surpasses a trivial baseline and avoids overfitting while ensuring robust performance across all categories. Specifically, the model must minimize misclassification of unrelated topics, with some tolerance for errors between closely related classes (e.g., “artificial intelligence” and “programming”). Additionally, the project aims to identify an appropriate scalar performance metric to comprehensively evaluate the algorithm’s overall effectiveness.
