About me

Comprehensive Time Series Application

An application developed to be able to create a large variety of time series models based upon time series data submitted by the user. So far this app works on annual, quarterly, monthly, and daily data.

Some of the models which can be developed include Summary models, Basic Time Series Models, Linear Regression Models, Exponential Smoothing models, and both seasonal and non-seasonal ARIMA models. The application also provides model statistics – which can be used for judging the best model – as well as the forecasted values of each model of your choice.

You can try the app out for yourself here!

If you need data for the app, click me! Please note that this data is Quarterly in nature, and needs to be converted to tsibble in the app. Follow Step 3 of the instructions. If you have your own data, it should work just as well!

• Project Source: Undergraduate Program
• Project type: Single-Person Project
• Class (if any): BAS 475 - Time Series Analysis
• Skills / Used: Shiny, R, Time Series
• Libraries Used: ggplot / plotly, fpp3

Close Project

Movie Recommendation engine

A shiny app which I built which uses association rule mining to determine and recommend movies that users may like based on an input of their favorite movies. Instructions on how to use are provided on the opening screen.

You can try the app out for yourself here!

• Project Source: Undergraduate Program
• Project type: Single-Person Project
• Class (if any): BAS 479 - Analytical Capstone
• Skills / Used: R, Data mining, Shiny, Market Basket Analysis
• Libraries Used: Shiny, arules

Close Project

FirstCare Medical Ceneter database

As part of my Undergraduate Database class, we were tasked with creating a fully functioning working database which can generate reports with a fully funcitoning working user interface. The database was first outlined in an ERD developed using Microsoft Visio, and then built in Microsoft SQL server. This database was then imported into Microsoft Access, which is what was used to create the User Interface.

This user interface includes ways to submit new data to every database entity, as well as form request interfaces to make querying the database more efficient and user friendly.

You can download the full report, which includes all the code for the database as well as graphics detailing the user interface here.

• Project Source: Undergraduate Program
• Project type: Group Project
• Class (if any): IM 342 - Database Systems
• Skills / Tools Used: SQL, Microsoft Access
• Collaborators: Justin Ketterl, Logan Tennel

Close Project

Where Should You Live?

A tool me and my team built to assist users in finding the best place for them to live based on a large variety of different data. Using the tool, users can decide for themselves what matters the most for them, and find a state / city that has the most factors in the areas that are important to them.

The data for this tool is pulled from a variety of sources, and is referenced in the website itself. We used a K-Nearest Neighbors (KNN) model to find similar cities to the one that was selected by the user. More information on the process of creating this KNN model as well as instructions of how to use the tool is included in the link below.

If you would like to try to tool out yourself, click here!

• Project Source: Georgia Tech MSA Program
• Project type: Group Project
• Class (if any): CSE 6242 - Data and Visual Analytics (DVA)
• Skills / Tools Used: Data: Pandas, Numpy, SkLearn // Website: HTML, CSS, Javacsript (D3)
• Collaborators: Josh Cantera, Andrew Crawford, Sam Courson, Mark Ramirez

Close Project

Flight Traffic Analysis

A Tableau project used to analyze the cancellations of flights over a 5 year period. All data for this project was downloaded from the Bureau of Transportation Statistics. Data cleaning was conducted in both R and Python, and all visualizations were made using Tableau. Some of the visualizations made include the following:

• An animation showing the number of cancelations, as well as the cancelation rate, of all airports in the United States over time.
• A summary of cancelations across all major airlines and by cancelation reason.
• An analysis of cancelations between major airports (seen above) which allows users to examine cancelation rate between individual airports.

• Project Source: Georgia Tech MSA Program
• Project type: Group Project
• Class (if any): CS 6730 - Visualization Principles
• Skills / Tools Used: Data: R and Python // Visualizations: Tableau
• Collaborators: Jared Coffsky, Christina York, Angela Zheng

Close Project

Best Buy Case competiton

Low selling SKU prediction engine.

As part of Project Week at Georgia Tech, Best Buy presented a case to all MSA students looking for a forecast model that they could use to predict the next week of sales for tens of thousands of low-selling SKUs. Whichever model we submitted needed to be able to predict sales precisely, while also running relatively quickly across the entire bredth of SKUs.

Our team created a Regression-Based Random Forest model utilizing lags and diferences, similar to time series models, in order to predict the sales of these SKUs. Overall, the model created by me and my team performed significantly better than those submitted by other teams of masters students, and we won the case competition.

Our model is now being used by Best Buy to solve this issue they have been having – to better predict the sales of all of their low-selling SKUs.

• Project Source: Georgia Tech MSA Program / Best Buy
• Project type: Group Project
• Skills / Tools Used: Data: Python // Visualizations: Tableau
• Collaborators: Joshua Cantera, Anamik Jhunjhunwala, Nikolaos Kavouras

Close Project

Bird Image Classification

Classifying images of birds through CNNs

In order to challenge ourselves for one of our major class projects, my team and I decided to go above and beyond and create an image classification model for local birds of Georgia. We used both normal CNNs and a Semi-supervised approach towards CNNs to accomplish this goal.

We attempted many different models, but the best we found was the Semi-Supervised models. In these, multiple "loops" are run, where in each a Convolutional Neural Network is run on a small subset of labeled data, it is then used to predict the labels on a much larger set of unlabeled data, and those predictions with an accuracy greater than a certain threshold are then added to the training set.

This allows us to create a more iterative model than we otherwise would have with just standard CNNs, and is a great way to build models like these in "the real world."

You can download the full report which details our findings here.

• Project Source: Class / ISYE 6740
• Project type: Group Project
• Skills / Tools Used: Python
• Collaborators: Mark Ramirez, Andrew Crawford, Samuel Courson

Close Project

Session-Based Recommendation Systems

Recommendation Systems using Session-based approaches

Recommendation systems are everywhere in today's world, and there are many different ways to implement such systems.

As part of this project, my team and I wanted to explore two different approaches in session based recommendation systems: implicit context recommendation systems (ISCON), and Heterogenous Global Graph Neural Networks (HG-GNNs)

Both of these methodologies were chosen as they represent the forefront of session based recommendation systems, which utilize "sessions" – essentially windows of use - as a factor to determine which products / items to recommend. Both ISCON and HG-GNN methods complete this, but with different structures of their actual implementations.

These analyses were conducted on two different datasets: the LastFM and Reddit datasets. Both datasets were chosen as they are incredibly popular for this type of analysis.

You can download the full report which details our findings here.

• Project Source: Class | CS 6240
• Project type: Group Project
• Skills / Tools Used: Python
• Collaborators: Abhinav Arun, Mehul Soni, Anshit Verma

Close Project

Society of Critical Care Medicine Case Competition

Exploring correlations between survival rate and native english speakers

It has been a common belief in the medical field that patients who speak little to no english often have worse experiences in critical care situations in the United States. My team sought out to see if this was actually reflected in the data during the 2023 SCCM datathon.

As part of the competition, we had access to the MIMIC-III dataset -- one of the largest and most well known datasets of emergency room patients. This dataset has just about every piece of information you could want about patients. However, it presents some unique problems.

One of the major problems we faced was the deliberate obfuscation of patient information within the dataset. In order to meet proper medical guidelines, data must be heavily obscured so as there is no identifying information whatsoever within the dataset, as well as obscuring different key parts of the dataset, such as dates. This means we had to spend a significant amount of time cleaning our data.

The other major problem we faced as part of this process was the lack of time to conduct any major analysis. The datathon itself consisted of 1.5 days of total time, meaning that whatever model we ended up settling on for our analysis would have to be one that runs fairly quickly.

Utilizing the guidance of some of the medical doctors on my team, I was able to develop a survival model in order to see if there is a difference between native english speakers and those who are not in number of doctor check ins before passing away while in the hostpial. Overall, we found that there was no major difference between the two populations in both time it takes to die, percentage of the population which pass on, or number of check ins before passing away.

As a result of the performance of our model and our unique findings, we won the datathon!

• Project Source: Society For Critical Care Medicine (SCCM)
• Project type: Datathon
• Skills / Tools Used: Python, Survival Models

Close Project