【Data Preprocessing Screen】 #
This screen allows you to perform data scaling. Data scaling is performed using the following formula. The values of ‘a’ and ‘b’ vary depending on the scaling method.
In=(Io-a)b
In: Scaled values
Io: Original values
1. Standardization:
・Method: Scaling to have a mean of 0 and a standard deviation of 1
・a: Mean of Io
・b: Standard deviation of Io
2. Max-Min Normalization:
・Method: Scaling the data to a range between 0 and 1
・a: Minimum value of Io
・b: Maximum value of Io – minimum value of Io
3. Abs-max Normalization:
・Method: Scaling the data to have a maximum absolute value of 1 while keeping 0 as 0
・a: 0
・b: Maximum absolute value of Io
4. None:
・Method: No scaling applied; original values are used as-is
【AI Training Screen】 #
To start training the AI, adjustments can be made to the parameters that define the structure of the neural network. Below, you will find an explanation of each parameter:
・Number of iterations:
This determines the number of models to be created. In neural network models, the initial network weights are assigned randomly. Consequently, even with the same hyperparameters, the accuracy of the model may vary each time it’s created. By creating multiple models, you can obtain the average error of the neural network under a given set of hyperparameters. This also allows for ensemble analysis using multiple models.
・Number of outputs:
The number of output parameters (non-editable).
・Number of epochs:
The number of times the training data is repeatedly used for learning. A higher value generally results in lower prediction errors on the training data, but excessively high values may lead to overfitting.
・Number of hidden layers:
The number of hidden layers in the neural network.
・Number of neurons in hidden layers:
The number of neurons in the hidden layers.
・Dropout of input layer:
A regularization process that probabilistically ignores the outputs of the input layer during training.
・Dropout of other layers:
A regularization process that probabilistically ignores the outputs of layers other than the input layer, during training.
・Size of training data:
The size of the data used for AI training when constructing a machine learning model.
・Size of batch:
Number of training examples processed in each forward and backward pass during training.
・Number of patience:
The maximum consecutive epochs with no improvement before early stopping, which prevents overfitting.
・Number of split:
The proportion of data to split for early stopping judgment.
・Batch normalization:
A process that normalizes the input to each layer of the neural network.
・Activation function except output layer:
The activation function for layers other than the output layer, which converts the sum of input values to the output value for each node.
・Activation function for the output layer:
The activation function for the output layer, which converts the sum of input values to the output value for each node in the output layer.
・Check random:
A process that shuffles the training data randomly before AI training. This ensures that different training data are used each time, preventing bias in validation data errors when creating multiple models.
・Loss function:
The function used to measure the error between predicted values of the model and actual values.
・Optimizer:
The optimization process that updates the weights and biases of each node to minimize the error between predicted and actual values.
・Number of epochs when learning stopped:
The number of times the training data is repeatedly used, before learning is stopped.
・Cumulated distance between validation data:
Cumulative distance between validation data points. A higher value indicates greater diversity in the validation data.
・RMSE after preprocessing:
Root Mean Square Error (RMSE) on the scale of the preprocessed data.
・RMSE before preprocessing:
RMSE when the data is reverted to its original scale before preprocessing.
・Relative error:
Relative error calculated as
Σ|(Actual Value – Predicted Value) / Predicted Value| / Number of Predicted Data
in Multi-Sigma.
・Correlation between prediction and actual:
Correlation coefficient between predicted values and actual values. Values close to 1 indicate a positive correlation, close to -1 indicate a negative correlation, and close to 0 indicate no correlation.
【AI Prediction Screen】 #
In this “AI Prediction Screen,” you can predict the output for specified input data. The AI model you’ve previously trained will be used to generate these predictions.
・ Ensemble Model:
This feature calculates the average output from multiple different models, allowing for more robust and accurate predictions.
【Optimization Screen】 #
In this “Optimization Screen,” you can use the AI model created through AI training and a multi-objective genetic algorithm to search for the optimal input conditions that result in the best possible output.
・ Size of generation:
The number of generations for exploring optimal solutions.
・ Size of population:
The number of individuals in each generation.
・ Number of crossover rate:
The proportion of pairs of individuals with favorable solutions (parents) that exchange genetic information (crossover).
・ Number of mutation rate:
The proportion of favorable individuals in each generation that undergo sudden genetic changes (mutation).
・ Ratio of elite:
The proportion of outstanding individuals in each generation that will be preserved and reported as the optimization results.
・ Scale factor:
A parameter that determines the strength of eliminating individuals with similar output values to maintain diversity within the population. A larger value leads to a wider range of eliminating individuals with similar output values.
