what is the target value mean?

The target value in an ANN is the desired output that you want your network to predict or approximate. It is the "ground truth" or the actual outcome that corresponds to a given set of input data. For example, if you are training a network to predict house prices based on features like size, location, and number of rooms, the target values would be the actual prices of the houses in your training dataset.The reason it is different from your input value is that the inputs are the variables you use to predict the target. They are the features or independent variables of your dataset, while the target is the dependent variable that you are trying to predict.
The target output value is essentially the same as the target value; it's the value that your model is attempting to predict for each input. In the context of the regression plots you mentioned, the target output values are plotted against the actual outputs produced by your ANN to assess the model's performance. A perfect model would result in a plot where all points lie on a 45-degree line, indicating that the predicted values match the target values exactly.
The target value is not something you calculate; it's something that should be part of your dataset. When you collect or prepare your data for training an ANN, you should have a set of input features (the variables you'll use for predictions) and corresponding target values (the outcomes you're trying to predict). If you're working on a supervised learning problem, the target values come from your dataset. For instance, if you have historical data with known outcomes, those known outcomes are your target values. You use this data to train your ANN so that it can learn to predict the target values from the input features. After training, you can then use the ANN to predict target values for new, unseen input data. If, however, you're referring to how the ANN calculates its output (which is an approximation of the target value) during training or prediction, this is done through a forward pass in the network, where the input data is fed through the layers of the network, and the ANN uses its learned weights and biases to calculate its output. The network's performance is then evaluated by comparing this calculated output to the actual target values using a loss function, which the training process aims to minimize.