use of neural nets for erosion prediction.

Will soon be sharing an update on how I used a simple Artificial Neural Network (ANN) architecture to predict beach vulnerability.

This is how the sensitivity analysis was conducted to tune hyperparameters:

---
def NN_reg_mod(layers,neurons, learning_rate,patience,dropout,batch):
split_ratio_v = 0.90
split_ratio_ts = 0.90

temp_prof = all_storms
learning_rate = learning_rate
    
X, y = temp_prof.loc[:,['duration_H', 'surge_sum', 'wave_R',
                         'sl_initial', 'Hs_max']], temp_prof.loc[:,'beach_change':'beach_change']

#x_train, y_train = X[:int(len(X)*split_ratio_ts)], pd.DataFrame(y[:int(len(y)*split_ratio_ts)])
#x_test, y_test = X[int(len(X)*split_ratio_ts):], pd.DataFrame(y[int(len(y)*split_ratio_ts):])

x_train, x_test,y_train,y_test = train_test_split(X,y,test_size = 0.2,random_state = 42)

x_train_scaled, x_test_scaled = scale_datasets(x_train, x_test)

es = EarlyStopping(monitor='val_loss', mode='min', verbose=1, patience=patience)


#Creating model using the Sequential in tensorflow
model = Sequential()
for i in np.arange(1, layers+1,1):
    model.add(Dense((neurons[i-1]), kernel_initializer='normal', activation='relu'))
    model.add(Dropout(dropout))
Dense(1, kernel_initializer='normal', activation='linear')

msle = MeanSquaredError()
model.compile(loss=msle,optimizer=Adam(learning_rate=learning_rate),metrics=[msle])
#train the model
history = model.fit(x_train_scaled.values,y_train.values,epochs=10000,batch_size=4,
                    validation_split=(1-split_ratio_v), verbose=0, callbacks=[es])
train_accuracy = r2_score(y_train.beach_change,model.predict(x_train_scaled)[:,0].flatten())
test_accuracy = r2_score(y_test.beach_change,model.predict(x_test_scaled)[:,0].flatten())

return(train_accuracy,test_accuracy)
---