Projects
Python | C++ | MATLAB
Stock Price Predictor - Fall 2024
COE 379L (Intro to Machine Learning and Data Science) Final Project
Advisor: Dr. Tan Bui-Thanh
-
Compared Long Short-Term Memory (LSTM) recurrent neural network predictions with those of standard fully connected neural networks for predictions of 1 day (with or without sentiment data taken from NewsAPI for the particular stock) or 20 days of stock prices for a particular company(all with different lookback windows). Longer predictions (20 days) use the past 180 days of stock prices as input, while shorter day-to-day predictions have a lookback window of 3-5 days (all parameters can be user-varied).
-
Model optionally uses a wavelet function (DMD) to smooth the price data and decrease prediction error
-
Also optionally uses a Hankel Matrix decomposition of stock prices (existing method, different application) and Black Scholes Option price as new features.
-
Achieved minimum errors (mean absolute percent error) of 5.375% for long-term predictions and <1% for short-term predictions (measured across 10 different stocks).
Prediction/Graph Example
Conway Game of Life
Built a program for the Conway Game of Life in C++. The program outputs the board to the computer screen after each iteration of the game sequence.
Cardiovascular Disease App - Spring 2023
COE 332 (Software Engineering and Design) Final Project
-
The project is a containerized Flask-Redis-Kubernetes application that analyzes behavioral risk factors for cardiovascular diseases across the United States of America.
-
The app includes:
-
several endpoints for retrieving specific data subsets, such as by class (cardiovascular rates or risk factors), topic (specific cardiovascular diseases or risk factors), location (states), and breakout category (gender, age, or race)
-
functionality to post, delete, and retrieve data from a Redis database
-
various data analyses (calculating most affected populations to specific diseases).
-
visualizations in graphs (of correlations between behavioral patterns like smoking and different cardiovascular diseases like stroke).
-
Graph Examples
Age standardized rates for CHD and Risk Factors
Age-standardized Detrended CHD Rates and Risk Factors
Example Analysis
Correlation coefficients between 3 risk factors and CHD in Texas:
{ "Correlation coefficient between consuming fruits and vegetables less than 5 times per day and coronary heart disease": "Not enough data", "Correlation coefficient between obesity and coronary heart disease": 0.38723599856793467, "Correlation coefficient between physical inactivity and coronary heart disease": 0.7849728887395832} }
Detrended CHD Rates and Risk Factors
Traveling Salesman Problem - Fall 2022
COE 322 (Scientific Computation) Final Project
-
Explored various optimization strategies for Amazon delivery truck drivers: 2 local search strategies, the greedy-route and the 2-opt, and a more effective combination of the greedy-route and the 2-opt algorithm.
-
The program (written in C++) optimizes delivery routes for one or two drivers. A cost parameter, which includes fuel price, driver’s minimum wage, and other time-related costs (Amazon Prime Customers get priority), is minimized. The model is dynamic: random new addresses (in number of addresses, locations, frequency, and prime/non-prime) can be added after the driver(s) have begun their route.
-
Algorithm and time-calculation performance for each of the three search strategies, various data sizes (number of delivery addresses), number of drivers, and model dynamism was also calculated.
-
Algorithm optimization can lead to over a 60% reduction in cost. (View Example)
Example:
Random Initial Route, Cost = $2506.19
Most Optimized Route, Cost = $972.19
Individual Attentional Psychology Experiment
Individual Research w/ MATLAB PsychToolbox
Developed a touch-screen based experiment intended for attention study. Experiment assigns points based on participants' reaction times (RT) to choosing trial-specific targets amidst distractors including salient (large and moving) & non-salient distractors. Image choices also include high-frequency (repeat across multiple trials) & high-value targets (less frequent, more point value targets). Data is saved to Excel files (image touched, RT, points per touch, etc.) and as text files for the total points scored per participant.
Group Python Projects
-
Tower Build
-
Worked in a team of 3 to build a cost-effective index card tower. The requirements were to build an at least 36 inch tower able to hold a stack of pennies (10 pennies). Over 2 weeks, we followed the Engineering Design Cycle to construct a tower surpassing requirements.
-
Used Python to optimize structure and maximize profit. Imported turtle to sketch design and CAD to model structure.
-
-
Connect 4 Game
-
Built a game that displays instructions for connect 4, asks players for their names & colors, allows them to play with key presses, and then records the winner in "hall of winners" document.
-
Used libraries numpy and pygame.
-
-
Go Game
-
Built a simple Go game (10x10 board) displayed in console with NumPy.
-
Counts number of moves and keeps score for each player.
-
