Click image to view project
Project Summary
Unsupervised Weather Analysis
Project that uses unsupervised models like clustering to find similarities of climate change among 16, selected U.S. cities. Algorithms also used: Jaccard similarity, Hamming distance and time series.
College Football Predictions
Used several prediction models including Logistic Regression, GLM Regression (Binomial) and Naive Bayes. Results measured with confusion matrices on train/test data sets. Full EDA on complied data downloaded from covers.com
Temperature Predictions
Facebook Prophet has been shown to be a superior modeler over Statsmodels ARIMA for forecasting and identifying trends in large time series data.
(short Video)
(long video)
Simulated Kaggle Entry
Project crafted for Kaggle which results finished 8th among 44 participants. Focus on feature selection using Boruta and BorutaShap as well variable inflation factor (VIF) and standard deviation checks. SMOTE and balance data algorithms enforce proportional values for desired features. Prediction models include XGBoost, decision tree, random forest, logitstic regression. Visualizations with UMAP. Jupyter Notebook runs on Ubuntu 22.4
Predicting Tropical Systems from Satellite Images
Using CNN models, the project aims at detecting tropical systems from satellite images archived at nhc.noaa.gov. Due to the limited number of images available, the code replicates images by augmenting them.
(short Video)
(long video)
(presentation slides)