Exploring LSTM, GRU and Convolution with NOAA buoy data
This project uses time series prediction to create a model capable of predicting water temperature. "This project employs both single and multivariate time series data to develop models utilizing Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Convolutional Neural Network (CNN) architectures. I perform exploratory data analysis on the data, handling missing values and renaming columns for clarity. I also performed some feature engineering by creating new time-related features to capture the cyclical nature of a 24 hour day. I utilized sine and cosine functions to encode the cyclical nature of daily time patterns, aiding the model's understanding of time-related dependencies. I performed single variable prediction and multivariate time series forecasting on the data using different LSTM, GRU and Convolutional models. "The notebook also includes observations on the parameter count for each model, highlighting the differences in their complexities. The notebook includes comparisons between single-variable and multivariate models, exploring how different input features and model architectures affect prediction accuracy.