Sentiment-Driven Stock Market Prediction using Twitter Data (Current Project)
Currently developing an MLOps pipeline for real-time sentiment analysis of Twitter data.
Extracting sentiment-related features using natural language processing (NLP) techniques (BERT/VADER), integrated with stock price trends, and training predictive models for stock market performance.
Energy Consumption Prediction of Residential Buildings using Machine Learning [Manuscript]
Collaborated with a team to analyze energy benchmarking datasets from eight U.S. metropolitan areas.
Conducted a comprehensive study utilizing extreme gradient boosting (XGBoost), random forests (RF), and artificial neural networks (ANN) to predict energy consumption in residential buildings.
An Attention-Based Deep Learning Model of Clinical Events in the Intensive Care Unit
Used the MIMIC-III dataset, which consists of anonymized data from over 40,000 patients.
Collaborated with a team to develop a deep learning model utilizing gated recurrent units (GRUs) and an attention mechanism, a transformer-like mechanism, to predict clinical events.
Estimating Energy Consumption in Residential Buildings in the City of Chicago [Manuscript]
Outstanding project of the quarter in CS230 Deep Learning at Stanford University.
Used the 2018 Chicago Energy Benchmarking dataset and implemented various machine learning models, including support vector machines (SVM), RF, and ANN, and conducted a comparative study.
Exploring Better Ways to Segment Lecture Videos Based on Topic Transition
Contributed as a team member to develop a model for segmenting lecture videos.
Utilized probabilistic latent semantic analysis (PLSA) as a generative approach to model text data and analyze lecture transcripts to identify topic transitions.
Partnered with a team to develop a three-invariant continuum cap plasticity model for large soil deformations, capturing nonlinear shear, compaction, and tensile behaviors under various loading conditions​.
The proposed model incorporates viscoplastic regularization to address rate-dependent effects and improve stability during softening, enabling accurate simulations of soil mechanics like triaxial compression and shear band formation.
Built a computational framework using the enhanced finite element method (EFEM), which is capable of modeling geomaterial failure caused by strain localization such as fracture, cracks, slip lines, and shear bands.
Investigated plasticity- and damage-like softening models to capture failure response in brittle geomaterials.
Cooperated with a group to design a 3D dynamic modeling method for powder forming processes using an efficient node-to-surface contact algorithm for evaluating density distribution during powder compaction.
The method incorporates large finite element deformations and frictional contact modeling, improving accuracy in simulating powder compaction and predicting non-uniform density distribution.
Participated as a key team member in implementing a computational algorithm for the dynamic simulation of powder compaction processes using a node-to-segment contact model and a double-surface cap plasticity theory.
The algorithm efficiently handles large frictional contact deformations, accurately simulating stress and density distributions in powder die-pressing with various component geometries​.
Modeling Contact (3-Dimensional and Static Analysis) [Manuscript]
Played a collaborative role in a team to model 3D large plastic deformation in powder forming using a node-to-surface contact algorithm with penalty and augmented-Lagrange methods for frictional contact simulation​.
The model successfully predicts non-uniform density and stress distributions, demonstrating efficiency through practical examples like deep drawing, tablet pressing, and shaped-charge liner compaction.
