Optimasi Backpropagation Menggunakan Grid Search untuk Penentuan Penerima Beasiswa Bidik Misi Berdasarkan Biodata Pendaftaran Mahasiswa Baru

Backpropagation Optimization Using Grid Search to Determine Bidik Misi Scholarship Recipients Based on New Student Registration Biodata

Abstract

The Bidik Misi Scholarship Program is a form of educational assistance provided to outstanding students from economically disadvantaged families. This program aims to expand access to higher education, reduce the gap in learning opportunities, and support the improvement of human resource quality. In practice, the selection process for scholarship recipients is still often carried out manually based on assessments of various administrative, social, and economic criteria. This process has the potential to create subjectivity, requires a relatively long time, and increases the possibility of errors in decision-making. Therefore, a prediction system is needed that can assist the selection process more objectively, quickly, and accurately. This study developed a prediction model for Bidik Misi Scholarship recipients using the Backpropagation algorithm optimized through the Grid Search method to obtain the best hyperparameter combination. The data used were new student registration biodata covering various social, economic, and academic indicators. The model was built using an Artificial Neural Network by comparing the performance of the Grid Search optimization model and the standard Backpropagation model. The results showed that the optimized model using Grid Search achieved an accuracy rate of 88.46%.

In comparison, the standard Backpropagation model with the Stochastic Gradient Descent (SGD) solver produced a higher accuracy of 96.15%. These findings demonstrate that selecting the right hyperparameters significantly impacts model performance and indicate that the standard Backpropagation model is more appropriate for the characteristics of the data used. The resulting system is expected to support the scholarship recipient selection process more efficiently, objectively, consistently, and transparently.