βž•Dataset and Parameter Optimization

70% of the dataset is used as the training set, while the remaining 30% is reserved as the testing set and is not involved in the model training process. Furthermore, within the testing dataset, we have also allocated 30% of the training data as a validation split while training the model, to help the model adjust its weights in a more generalized manner and avoid overfitting during training. The model is trained with the objective of minimizing the Mean Absolute Error as the central metric. The model is trained iteratively by varying different parameters within the model. Learning rate, activation function, number of CNN layers, number of filters, number of LSTM layers, and number of neurons in each layer have been experimented with to optimize the model architecture (Figure 5). The bold values indicate the values used that are considered optimal for a specific model. It is worth noting that changing the number of CNN layers in the model slightly improved accuracy, but not significantly enough to justify the additional computational cost. Similarly, with the number of filters. The parameter optimization settings can be seen in Table 2, and the training and testing error metrics can be observed in Tables 3 and 4, respectively.

Table 2. Parameter optimization value setting

ParametersValues

Learning rate

0.0001, 0.0005, 0.001, 0.005, 0.1

Activation function

ReLu, Tanh

Activation function

1,2,3,4

CNN Layers

1,2,3

No. neurons in first layer of LSTM and TDL

128,256,512

No. filters

1,2,4,8…32

Table 3. Training error metrics

MetricValues

Train R-Square

0.985

Train Mean Absolute Error (MAE)

167.558

Train Mean Absolute Percentage Error (MAPE)

0.0224

Train root mean square error (RMSE)

199.036

Table 4. Testing error metrics

MetricValues

Train R-Square

0.941

Train Mean Absolute Error (MAE)

242.408

Train Mean Absolute Percentage Error (MAPE)

0.031

Train root mean square error (RMSE)

413.912

Last updated