Based on data, the open unemployment rate according to the highest education graduate in Indonesia shows the number of semester unemployment which has an unstable value, sometimes up and sometimes down. This study aims to implement the ability and performance of one of the training functions on the backpropagation algorithm, namely one-step secant, which can later be used as a reference in terms of data forecasting. The one-step secant algorithm is an algorithm that is able to train any network as long as the input, weight and transfer functions have derivative functions and this algorithm is able to make training more efficient because it does not require a very long time. The data used in this study is open unemployment data according to the highest education completed in Indonesia in 2006-2021 based on semester, which is sourced from the Indonesian Central Statistics Agency. Based on this data, a network architecture model will be formed and determined using the One-step secant method, including 14-13-2, 14-16-2, 14-19-2, 14-55-2, and 14-77- 2. From these 5 models, after training and testing, the results show that the best architectural model is 14-19-2 (14 is the input layer, 19 is the number of neurons in the hidden layer and 2 is the output layer). The accuracy level of the architectural model for semester 1 and semester 2 is 75% with MSE values of 0.00130797 and 0.00388535.