Course assessment model in vocational education based on BPNN optimized by genetic algorithm

Why measuring skills in new ways matters

As more young people in China and around the world choose vocational colleges, schools are under pressure to judge students fairly and quickly. Traditional course assessments often rely on teacher impressions or simple averages of test scores, which can be subjective, slow to update, and hard to scale to thousands of learners. This study explores how a tailored form of artificial intelligence can turn messy records from many sources into clear, reliable pictures of student performance, helping schools support learners and improve teaching quality.

The challenge of judging real-world learning

Vocational education focuses on practical skills: welding a joint, wiring a circuit, or running a business simulation. Assessing such learning is harder than grading a single exam. Colleges collect many pieces of information—attendance, lab work, teacher ratings, and more—often stored in different formats. Human evaluators may weigh these factors differently, leading to inconsistent results. At the same time, some modern AI models need huge datasets and computing power, which many schools do not have. The authors set out to design a lighter, more transparent system that could handle diverse data while keeping judgments as objective as possible.

A smart engine tuned for vocational data

The heart of the study is a course assessment model built on a backpropagation neural network, a classic AI technique that learns patterns by repeatedly comparing its predictions with actual outcomes. To make this network work well on educational data, the researchers first clean and scale all input values so that different indicators—such as test scores or attendance rates—can be compared on the same footing. They then use an "entropy" method to calculate how informative each indicator is, giving more weight to factors that vary meaningfully across students and less to those that change little. This reduces the influence of guesswork about which indicators matter most.

Borrowing from evolution to improve learning

On top of this core network, the team adds a genetic algorithm, inspired by natural selection. Possible settings for the network’s internal connections are treated like individuals in a population. Through repeated rounds of selection and recombination, the algorithm searches widely for promising combinations of weights and thresholds, helping the system avoid getting stuck in poor solutions. Once a good region is found, a second method, known as Levenberg–Marquardt, takes over to fine-tune the parameters quickly and precisely. This two-stage process balances broad exploration with fast, local adjustment, allowing the model to learn accurately from relatively modest datasets.

How well does the model work in the classroom?

To test their approach, the authors built two real-world datasets from Chinese vocational institutions, containing thousands of course records from multiple majors and schools, as well as additional public student datasets. They compared their hybrid model with several alternatives, including traditional neural networks, recurrent networks, and other optimization schemes. Across measures such as prediction error, stability over repeated trials, and alignment between predicted and actual course scores, the new model consistently came out ahead. It reduced typical prediction errors into the single-digit percentage range, achieved very high agreement with real assessment outcomes, and maintained strong performance when trained and tested on different student groups.

What this means for students and schools

Put in everyday terms, the study shows that a carefully designed AI system can turn complex, multi-faceted records of vocational learning into trustworthy course evaluations that are less biased and more consistent than many current methods. By automatically balancing knowledge tests, hands-on practice, and broader qualities like teamwork, the model mirrors the goals of modern vocational education while easing the workload on teachers and administrators. Although further work is needed to address issues such as fairness across different student groups and adaptation to other countries’ systems, this framework points toward smarter, more transparent ways to recognize practical skills and support learners on career-focused paths.

Citation: Luo, W., Zang, L., Liang, W. et al. Course assessment model in vocational education based on BPNN optimized by genetic algorithm. Sci Rep 16, 12686 (2026). https://doi.org/10.1038/s41598-026-43397-9

Keywords: vocational education, course assessment, neural networks, genetic algorithms, educational data mining