This repository contains the data and code for the following research paper:
Predicting chlorophyll-a concentrations in the world's largest lakes using Kolmogorov-Arnold Networks (doi.org/10.1021/acs.est.4c11113)

This GUI is a general-purpose tool for predicting and forecasting Chlorophyll-a (Chl-a) concentrations. Please note that this tool was NOT used to generate the results for our study, although it contains the same models and is intended for general application.

This tool provides a selection of univariate time series models to predict and forecast Chl-a concentrations. The included models are:

• KAN (Kolmogorov-Arnold Networks)
• MLP-NN (Multilayer Perceptron Neural Network)
• LSTM (Long Short-Term Memory)
• GRU (Gated Recurrent Units)
• RF (Random Forest)
• GPR (Gaussian Process Regressor)
• SVR (Support Vector Regressor)

• Lag Features: 12 lag features are automatically generated, based on the monthly frequency of the original Chl-a dataset. If your dataset has a different frequency, you can modify the lag duration in the code.
• Forecast Duration: The default forecast duration is set to 6, which corresponds to forecasting the next six months in the original dataset. If your data uses a different frequency, you are encouraged to modify the forecast duration accordingly in the code.

After selecting a model, the tool will train it using your dataset, and it will display a plot of actual vs. predicted values. Key metrics such as MAE, MSE, and R² will also be provided. You can then use the trained model to forecast future values.

Since this tool is designed for general use, performance may vary depending on your dataset. For optimal results, consider fine-tuning the hyperparameters specific to your data.

The following libraries are necessary to run the tool. These can be installed via standard package managers (e.g., pip) and are not custom or local libraries.

• tkinter
• pandas
• matplotlib
• sklearn
• numpy
• statsmodels
• torch
• KAN
• tqdm
• tensorflow
• deepkan
• scipy
• RandomForestRegressor
• GaussianProcessRegressor
• SVR

1. Ensure all required libraries are installed.
2. Load your Chl-a dataset into the tool.
3. Select a model, train it, and review the performance metrics.
4. Forecast future values based on your trained model.