Unlocking Success: Evaluating the Impact of Industry Education Integration Models on Applied Universities Using Back Propagation Neural Networks – Insights from Scientific Reports

BPNN: Selection and Construction

The Backpropagation Neural Network (BPNN) is a popular tool in the fields of pattern recognition, classification, and regression. It’s a multilayer feedforward neural network that learns through the backpropagation algorithm.

Its structure consists of an input layer, one or more hidden layers, and an output layer, all connected by weights. Training the BPNN involves two main steps. First, in the forward pass, input data is processed to produce an output. Second, based on the difference between the predicted and actual outcomes (the error), the network adjusts its weights by moving backward through the layers, aiming to reduce this error over time.

Each neuron’s output is calculated by taking a weighted sum of inputs from the previous layer and applying an activation function. The equations used for these computations help define how information flows through the network.

Next comes error calculation, where we determine the difference between the network’s output and the actual values. The BPNN uses this error to adjust the weights in a process called backpropagation. This involves spreading the error back through the network to each neuron and modifying the weights accordingly.

The learning process includes updating weights using a method called gradient descent, allowing for gradual improvement in predictions. This is crucial for the BPNN model’s effectiveness.

The BPNN used in this study is particularly designed to analyze factors related to student performance. The input layer incorporates various features like academic performance, participation in activities, and feedback from internships. These features are carefully chosen and normalized for consistency.

The model uses two hidden layers to manage the complexity of educational predictions. These layers allow the model to grasp both linear and nonlinear relationships in the data, which is essential for learning from the diverse nature of student data.

Weight initialization is important for model performance, and in this case, random normal distribution is used to help diversify feature learning. The Adam algorithm optimizes the training process, adapting the learning rate for faster convergence and better training efficiency.

Through experiments, the hidden layers were perfected, with the first containing 128 nodes and the second below 64. This setup balances the model’s learning capacity without risking overfitting, which is essential for maintaining accuracy and robustness in predictions.

The model’s structure, activation functions, and loss functions were chosen with care to achieve a reliable and efficient outcome. ReLU activation function was picked for hidden layers due to its effectiveness at improving convergence speed. The output layer used different functions depending on task requirements, ensuring suitable predictions for various tasks.

The mean squared error (MSE) was selected as the loss function, effectively guiding the model’s learning process. Additionally, the Adam optimization algorithm was utilized to improve weight updates and enhance training capabilities.

The following key components were part of the implementation of the industry-education integration model:

1. Intelligent Course Recommendation System: This system analyzes students’ data—like learning history and preferences—to recommend personalized courses.
2. Employment Guidance Services: This component helps students plan their careers based on their learning experiences and enhances their job market competitiveness.
3. School-Enterprise Cooperation Mechanism: This encourages collaboration between educational institutions and enterprises, promoting resource sharing and mutual benefits.

In the intelligent course recommendation system, multidimensional data is analyzed to suggest courses tailored to each student’s learning style and achievements.

For employment guidance, data preprocessing starts by collecting and cleaning students’ academic information. Next, essential features are extracted to guide BPNN in training and producing valuable career planning advice.

The school-enterprise cooperation mechanism establishes a platform for communication between universities and businesses. This platform supports project collaboration, ensuring that educational content aligns with industry needs.

Overall, the BPNN model plays a vital role in enhancing decision-making processes by predicting the best paths for students, from course selections to career opportunities. With continuous updates based on feedback and data flow, BPNN remains adaptable and effective in supporting educational and professional development.