In addition to neurons, each layer has bias. They are a certain number of constant values: b1, b2, ... bm, where m is the number of bias in the layer.
All these values are scaled by a specified transfer function such as tansig, logsig, purelin, etc. This is required to prevent output value becoming infinity or zero. So ultimately we get o number of values from the input layer if there are o neurons in the input layer. The output of the input layer is given as input to the first hidden layer. The hidden layer is also similar to the input layer. It also has a specified number of neurons, bias and a transfer function.
The output layer is also similar to the input layer and has a specified number of neurons, bias and a transfer function. Finally we get one output from the output layer. The neural network adjusts the neuron weights based on the output value obtained. It compares the output with the expected value and adjusts the weights to give more correct output using the training data.
The above step is repeated several times until the weights assigned produce the required output. Finally, when we give the input stock price, EMA, RSI and MACD data on any date, the output is close to the future date stock price. Below are 62 lines of the 1773 lines developed for forecasting, using neural network with technical analysis