I want to make a model to predict the land use of some fields but my accuracy after making my model is about 50%, which is so low. I want to improve the accuracy of my model to my case and as I am new in neural networks I need some help with it.
Here is the model I've made. I work with historic data from the fields, in CSV format. I use 4 columns of data to train the model ('mean_B04','mean_B08','NDVI','H_NDVI') and I have 9 different land uses (0,1,2,3,4,5,6,7,8) to classify the predictions.
#Data to train and to classify
df1 = df[['mean_B04','mean_B08','NDVI','H_NDVI']]
df2 = df['Num_uso']
Data_array = np.array(df1)
Uso_SP_array = np.array(df2)
#Split testing and training datasets
xTrain, xTest, yTrain, yTest = train_test_split(Data_array, Uso_SP_array, test_size=0.15, random_state=42)
model = Sequential()
model.add(Bidirectional(LSTM(100, return_sequences=True, activation='tanh'), input_shape=(Filas_entrada,Canales_entrada,)))
model.add(Bidirectional(LSTM(100, activation='tanh')))
model.add(Dropout(0.5)) model.add(Dense(50, activation='relu'))
model.add(Dense(9,activation='softmax'))
model.compile(loss='sparse_categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
#Run the model
model.fit(xTrain, yTrain, epochs=1, batch_size=24, validation_split=0.1)
#Get a list with the predictible classes
ListOfSPusos = df_Train['Num_uso']
SPUsos = ListOfSPusos.drop_duplicates()
SPUsos = SPUsos.sort_values()
SPUsos = SPUsos.reset_index(drop=True)
#Predict for test data
yTestPredicted = model.predict(xTest)
y_classes = SPUsos[np.argmax(yTestPredicted, axis = 1)]
