I worked on the California Housing Data dataset. The dataset provided 9 features that I used to try to predict the median housing data. I performed expanded exploratory data analysis in order to consider the best ways to encode categorical data and scale the numeric features. I also used PCA to add features, as well as K-means clustering to identify potential meaningful cluster identifiers. THis informaiton was fed into several regression models/estimators, all of which were then run through gridsearch and cross validation to determine the best model on split training and validation data.
Of the models considered, the best was was a Gradient Boosted Decision Tree with max depth set to 8 (with default parameters otherwise [e.g., learning rate = 0.1]). The score on validation data was ~0.84, which is a respectable score but there is likely room for improvement. It scored well on the training data (0.95), so it is unlikely that the model is overfitting.
Programming, Python, Statistics, Numpy, Pandas, Matplotlib, Scikit-learn, Dataframes, Data Modeling, EDA, Data Visualization, Data Reporting, Classification, Supervised ML, Cross Validation, Grid search, Unsupervised ML, PCA, Seaborn, Folium