Fast Artificial Neural Network

Summary

This tool uses the Python packages scikit-learn and NumPy to establish an artificial neural network regression model.

Usage

Navigate to and click on ALS Forest > Regression Analysis > Fast Artificial Neural Network.

Settings

• Import Training Data: Refer to Sample Data and Independent Variables.
• Independent Variables: Refer to Sample Data and Independent Variables.
• Momentum (default value is "0.6"): Set the momentum parameter in ANN regression analysis for selecting the optimized path.
• Learning Rate (default value is "0.7"): The global learning rate for training the network.
• Accuracy Assessment: Based on the K-Fold cross validation model, a sample would be partitioned into k subsets according to input K-Fold value (no less than 2). Take one of the subsets as a validation dataset and the remaining subsets as training datasets to form a model, then run this model and test the fitting of the validation set to training sets. Repeat this process until every subset is treated as a validation set at least once then select out the model with the least MSE (mean square error) as the optimal model.
• Save Regression Model: Tick the checkbox to save the ANN model (Fast Artificial Neural Network.model) under the output path.
• Save Regression Dataset: Tick the checkbox to save the ANN training dataset (Fast Artificial Neural Network.csv) in .csv format under the output path.
• Output Path: Choose an output directory. A fast artificial neural network regression model report (Fast Artificial Neural Network.html), recording the model's parameters and accuracy (R-square, RMSE), would be generated under this directory. A prediction result file (Fast Artificial Neural Network.tif), based on the fast artificial neural network regression model and input variables from a .tif or .csv file, would also be generated under this output directory.

Note: The dimension of imported sample/training data must be within the scope of independent variables, which may be adjusted accordingly. The model/result is based on the passed-in variables.