Projects
Modeling and Forecasting Ontario’s Monthly Energy Consumption
Live Quarto Document ↗ (opens in a new tab) | Github ↗ (opens in a new tab)
Forecasts for Ontario’s future monthly energy consumption were generated using a dynamic regression model that integrated temperature as an explanatory variable, achieving a 2% improvement in predictive accuracy over a baseline SARIMA model. Historical temperature and energy consumption data from 1997 to 2024 were analyzed, revealing key seasonal trends and non-linear relationships through time series decomposition to enhance model accuracy. A data-driven pipeline was established to produce reliable energy consumption forecasts, with the potential to inform infrastructure planning and optimize electricity production and storage.
Exploratory and comparative data analysis of Fertility Rates in South Korea
Live Quarto Document ↗ (opens in a new tab) | Github ↗ (opens in a new tab)
Conducted exploratory and comparative data analysis in R to examine fertility rate trends between South Korea and Japan, focusing on policies related to female employment, paternal leave, and child care availability. A predictive multi-linear regression model improved predictive accuracy by 10% over a baseline ARIMA(0,2,1) model, using rolling forecast origin cross-validation. Scenario-based forecasts revealed that increasing maternity leave proportions to 84.5% by 2030 could boost South Korea’s fertility rate by 20%, emphasizing the importance of government policies aimed at expanding maternal leave.
Coffee Bean Multi-Bean Classification | Live ↗ (opens in a new tab)
Evaluated two CNN models, Effnet B1 and Effnet V2 S, to classify four types of Arabica beans. Both models, trained on 300 images per class and tested on 100 images per class, achieved over 98% accuracy. Effnet B1 was preferred due to its smaller size (26MB) and shorter training time compared to Effnet V2 S (81MB). However, the models struggled to confidently distinguish between medium and dark roasted beans due to image shadows. Future improvements could focus on better differentiation and adding more bean types and shapes.
Visualizing S&P 500 | Live ↗ (opens in a new tab)
Utilized d3.JS for visualizing S&P 500 trends alongside FED Rate, NASDAQ, and KOSPI, presenting interactive graphs illustrating the compounded benefits of annual versus monthly investments in the S&P 500, emphasizing long-term growth compared to a savings account, and showcasing the potency of compound interest with a 368.3% outperformance of S&P 500 over a savings account.
AI Speechbot | Live ↗ (opens in a new tab)
AI chatbot on Node.js and Express framework that supports speech input for users facing typing or technology challenges. Potential to integrate with language models like OpenAI's GPT-4 through API support to enhance user engagement.
© Joe Lee.RSS