Unsupervised Classification of The Netherlands Landcover

Master programme course individual project

Hyperparameter-optimized Supervised Classification

Master programme course individual project

In this project, I was tasked to perform supervised classification to Land Use and Land Cover (LULC) from provided EuroSAT dataset using machine learning algorithm of my choice, Dask parallel computing, and incorporate hyperparameter optimization to find the best parameters for the task.

I used Decision Tree algorithm for the classification since I extracted only 4 features from the dataset (Mean, Median, Range, and NDVI index), assumed to be simple enough to not using more sophisticated algorithm such as Random Forest or ANN. Those statistical features were extracted using Rasterio and stored into Dask dataframe. Hyperparameter optimization was performed to find the best Decition Tree parameters combination for the task. The decision tree algorithm resulted in prediction scores after validation, feature importance visualized in histogram, and classification report for each LULC type.

Since the provided EuroSAT data is too big to be stored elsewhere, I could not make the code reproducible for everyone.

However, I can gladly showcase the Jupyter Notebook in action during live call or interview upon request.

Contact Rifqi

I am looking forward to connect and be in discussion with you