firstName

lastName

email

message

Got a project? Let's chat!

Alexandra Hernandez

C++ program that creates a graph using customized classes with basic properties as well as returning the shortest paths from parameterized nodes using different algorithms.

This program reads a provided `.csv` file, a creates a comprehensive graph using C++'s data structure library (`tuple` , `vector` , `queue` , etc.). Utilizing custom classes to implement the graph, it helps represent a complex network of relationships using node and edge management, eventually going beyond basic representation and implementing traversal algorithms to return the shortest defined paths. With algorithms such as Breadth-First Search (BFS), Depth-First Search (DFS), and Dijkstra's algorithm, each traversal returns the specific path based on the provided node. Additionally, the implementation incorporates disjoint set operations with union functionality, providing a robust toolkit for graph analysis and manipulation. Through carefully designed classes and methods, the program enables users to explore graph properties, perform sophisticated network traversals, and solve complex pathfinding problems with efficiency and precision.

Original picture is accessible [here](https://www.google.com/url?sa=i\&url=https%3A%2F%2Fpython.plainenglish.io%2Fgraph-theory-23468908321d\&psig=AOvVaw3fSJb5_xGU00FXbapEXVrz\&ust=1734233876001000\&source=images\&cd=vfe\&opi=89978449\&ved=0CBcQjhxqFwoTCICN2eOqpooDFQAAAAAdAAAAABAZ).


Graph Implementation Project

Analyzing and predicting California housing prices using census data from 2009-2020, alongside machine learning models such as Linear Regression and Random Forest to examine the differences in accuracy/variation, as well as predicted prices of a potential house in California.

Conducted an in-depth analysis of California housing prices by leveraging federal census data spanning from 2009 to 2020. We developed predictive models that explored the intricate factors influencing real estate valuations across the state by employing both Linear Regression and Random Forest modeling approaches. We examined critical variables including coastal proximity, average household income, and median house prices to assess their impact on housing market dynamics; research meticulously compared the accuracy and variation between these two modeling techniques, providing nuanced insights into the complex mechanisms driving California's housing market. The study not only offered a sophisticated statistical examination of price determinants but also demonstrated the potential for predictive modeling to generate precise estimates for potential house values in different regions of California.

Report is accessible [here](https://alexandrarh.github.io/cogs_137_reports/final_project.html).


Predicting Housing Prices in California

Compression/Decompression program using Huffman Tree encoding/decoding to turn input files into binary encrypted files, and decoding encrypted files using the saved Huffman Trees.

This C++ program implements a sophisticated file compression and decompression system built around a custom HCTree (Huffman Coding Tree) class. It utilizes C++ libraries like FancyInputStream and FancyOutputStream in its transformation of input files into efficiently encoded binary representations. The primary class of the programs, `HCTree` class serves as the core mechanism for encoding and decoding, enabling precise file compression that reduces data size while preserving original content integrity. Through carefully implemented functions within the `HCTree` class, the program can both compress files into compact binary formats and accurately decompress them back to their original state, demonstrating a robust approach to data handling and file transformation.


Huffman Compression/Decompression

Case study analyzing air pollution in the United States with tidymodels. Using data from EPA's pollution monitors, it examines factors such as poverty levels, road lengths, and education levels around monitor to determine the average levels of pollution per zip code.

Conducted an in-depth analysis of air pollution patterns across the United States using advanced machine learning techniques in R and RStudio. Drawing on comprehensive data from the EPA's nationwide pollution monitoring network, our group developed a sophisticated predictive model that examines the complex interplay of environmental and socioeconomic factors influencing air quality. By leveraging packages like `tidyverse`, `ggplot2`, and `tidymodels`, we employed linear regression modeling to investigate how characteristics such as road lengths, zip code demographics, education status, and poverty levels correlate with local pollution levels. Our study meticulously mapped and analyzed pollution concentrations. Through rigorous statistical analysis and tidymodels machine learning algorithms, we generated insights that illuminate the multifaceted nature of air pollution distribution and its relationship to community characteristics.

Report is accessible [here](https://alexandrarh.github.io/cogs_137_reports/cs02.html).


Case Study 02: Predicting Annual Air Pollution

Group case study analyzing researcher data on marijuana usage to determine what factors are best for marking recent usage of marijuana. Used metrics such as blood, oral fluid, and breath to examine marijuana dosages and compound differences.

Conducted a detailed case study examining marijuana usage detection methods using advanced statistical analysis in R and RStudio. Leveraging powerful data science packages like `tidyverse` and `ggplot2`, we systematically investigated multiple biological markers—including blood, oral fluid, and breath samples to identify the most effective indicators of recent marijuana consumption. By applying rigorous machine learning techniques such as Random Forest modeling, we evaluated the sensitivity and specificity of various marijuana compounds, with a specific focus on potential roadside detection applications for law enforcement. Our findings revealed a clear hierarchy of reliability, with blood samples emerging as the most dependable method for detecting recent marijuana use, closely followed by oral fluid measurements. This research provides critical insights into the complex landscape of marijuana detection methodologies, offering valuable guidance for forensic and medical professionals seeking accurate and timely assessment of recent cannabis consumption.

Due to research privacy protocol, code can't be displayed.


Case Study 01: Biomarkers of Recent Usage

Bridging the gap between older folks and mental healthcare with machine learning.

Done as a part of San Diego State University's [2024 Big Data Hackathon](https://bigdataforsandiego.github.io/), my teammate and I (Friar Health) created [MindTheGap](https://mind-the-age.streamlit.app/), a web app dedicated to helping elderly patients of SHARP Healthcare start their mental healthcare journeys by providing them with resources based on their information–both self-reported and medical staff provided. Utilizing a classification model trained on recognizing potential mental health "leanings" (via text-based classifications) and suggesting them resources based on those weightings. The goal of the app is to make starting to care for your mental health less intimidating, especially with the loads of information that need to be sifted through to find the right fit.

Source code can be accessed [here](https://github.com/alexandrarh/Team-133).


MindTheAge

Group project/special presentation for COGS 188. Discusses and demos the real world implications of racial biases in facial recognition algorithms.

Group project/special presentation done on the usage of Facial Recognition, and how internal–and external–biases such as racial biases are implicated in the algorithms in the real world, and what these implications can do for those utilizing such methods. Demo is done with Jupyter notebooks in Python, along with data from `utkcropped` and `age_gender` databases. By getting to the biases at their core, we then discuss and propose potential solutions in combating such biases from popping up in recognition algorithms.

Repository for the project can be found [here](https://github.com/alexandrarh/COGS188_group_project).


Facial Recognition and Racial Biases

Group final project for COGS 118B. Examines the relationship of location prediction from accent data labeling.

Examining differences/similarities between Accents from AccentDB database by using unsupervised machine learning algorithms via clustering techniques. Done all in Python and utilized its respective packages, and found that most ML algorithms had overlapping clustering, suggesting that overall accents and their similarities to one another are categorized similarly, yet could be differentiated into different (overlapping) clusters.

Source code can be found [here](https://github.com/alexavndra/cogs_118b_project/tree/main).


Accent Atlas

Web app that can generate battery cycle graphs from Arbin and Neware battery software.

Personal project-based [web app](https://battery-cycle-visualizer.streamlit.app/) that easily visualizes battery cycle data, using user-based parameters to visualize given cycles (single cycle, all cycles, or multiple certain). Takes in `.csv`, `.xls`, and `.xlsx` files, and allows for user to download cycle-specific data as a `.csv` file. Made with Python, Pandas, MatPlotLib, and Streamlit. This app is currently used by the Chen Battery Lab at the University of California, San Diego.

Source code can be found [here](https://github.com/alexavndra/battery-cycle-data-viz).


Battery Cycle Visualizer

Data science group research project that analyzes the relationship between caption keywords and DeviantArt content using Python.

Data science project that analyzes the relationship between DeviantArt content popularity and the keywords contained in their captions with Python, primarily in DeviantArt API, NumPy, Seaborn, MatPlotLib, and SciPy. Completed as part of COGS 108: Data Science in Practice class at the University of California, San Diego. When analyzing the relationship between caption keywords and popularity, we found to be a weak *positive* correlation between the two, suggesting that other factors such as follower count and overall visibility on the platform play a (possible) bigger role than captions themselves.

Source code can be found [here](https://github.com/alexavndra/digital-art-keywords/tree/main).


Projects