Rekomendasi Prioritas Penyelesaian Masalah Berdasarkan Notulensi Dari Bimbingan Skripsi Dengan Machine Learning

Recommendation of Problem-Solving Priorities Based on Thesis Supervision Minutes Using Machine Learning

Proses bimbingan skripsi sering kali menghasilkan notulensi yang memuat berbagai permasalahan yang harus diselesaikan dengan tingkat prioritas berbeda. Namun, penentuan tingkat prioritas tersebut sering kali subjektif dan memakan waktu. Penelitian ini bertujuan untuk mengembangkan sistem rekomendasi tingkat prioritas berdasarkan notulensi bimbingan skripsi menggunakan Machine Learning dengan algoritma SVM. Proses pengembangan model meliputi pengumpulan dataset notulensi dari mahasiswa semester akhir, augmentasi data melalui back-translation, preprocessing data, pencarian parameter optimal, serta pelatihan model SVM. Data penelitian diambil dari notulensi mahasiswa dengan total 1720 data yang telah diproses menjadi tiga kelas prioritas. Evaluasi model dilakukan menggunakan metrik akurasi, precision, recall, dan F1-score, menghasilkan akurasi tertinggi 97% dengan kernel RBF setelah penghapusan outlier. Sistem sederhana berbasis model SVM kemudian diuji 58 mahasiswa dari berbagai universitas untuk mengevaluasi tingkat kepuasan terhadap akurasi dan kemudahan sistem. Penilaian dilakukan melalui kuesioner berbasis skala likert. Hasil analisis menunjukkan bahwa pengguna memberikan penilaian positif terhadap akurasi hasil rekomendasi, kemudahan penggunaan, serta kejelasan informasi yang diberikan sistem.

The thesis supervision process often produces meeting minutes containing various issues that need to be resolved with different levels of priority. However, determining these priorities is often subjective and time-consuming. This study aims to develop priority recommendation system based on thesis supervision minutes using Machine Learning with the SVM algorithm. The model development process includes collecting dataset of meeting minutes from final-year students, augmenting data through back translation, preprocessing, finding optimal parameters, and training the SVM model. The research data consists of 1720 minutes that have been processed into three priority classes. The model evaluation uses metrics such as accuracy, precision, recall, and F1-score, achieving the highest accuracy of 97% using the RBF kernel after outlier removal. A simple system based on the SVM model was then tested with 58 students from various universities to evaluate user satisfaction regarding the system's accuracy and ease of use. The evaluation was conducted using a Likert-scale questionnaire. The analysis results indicate that users provided positive feedback on the accuracy of the recommendations, ease of use, and clarity of information provided by the system.