Abstract
In many countries, particularly in Iraq, the students' academic performance (SsAP) system is based on the final grade scores in high school. This final high school grade may not reflect the students' intelligence level or the interests that link the student to a relevant university. Also, skills are not used to predict score-related school or college. In this research, a seven-subject, one-grade, one-output (SOO) model was proposed to simulate the classic SsAP system to show that the predicting system is completely based on the previous year's score and not on the students' interests and skills. Moreover, a seven-subject, twelve-year, seven-output (STS) model, which used seven parallel deep neural networks with a scaled conjugate learning algorithm, was employed to determine the students' science, technology, engineering, art, and mathematics (STEAM) skills and interests across 12 grades and predict their corresponding most appropriate school. This article contributed to constructing two models: SOO model which simulates the classical Iraqi education system, and the STS model which predicts the acceptance of students according to the STEAM system, which is what makes it different from previous research. The results revealed that the SOO model properly simulated the classic SsAP system. Furthermore, the new approach based on STEAM education successfully predicted students' academic performance in line with their skills and interests over a twelve-year period. The overall accuracy rate of the two proposed models (SOO and STS) is about 99% with 10-5 histogram errors between the target and the actual output. However, the optimized epochs of the SOO model are 1000 epochs while the STS model got 10-600 epochs.