<![CDATA[Proses pemilihan program studi bagi calon mahasiswa baru, khususnya bagi mereka yang masih duduk di bangku SMA atau sederajat, merupakan salah satu momen pengambilan keputusan penting. Tak jarang pilihan yang salah berujung pada kegagalan studi atau kesulitan lain selepas menamatkan studi. Meski sudah mulai marak dilakukan di berbagai negara maju, sistem rekomendasi program studi berbasis machine learning untuk calon mahasiswa baru masih belum banyak dikembangkan di Indonesia. Penelitian ini dilakukan sebagai upaya rintisan sistem rekomendasi tersebut dengan menggunakan data pribadi dan akademik dari semua mahasiswa dan alumni program sarjana di Universitas Islam Indonesia (UII), di mana data prestasi akademik di masing-masing program studi digunakan sebagai ground truth label. Dari hasil penelitian ini, didapatkan sebuah model berbasis Random Forest (RF) dengan tingkat akurasi 86%, presisi 84%, recall 86%, dan AUC 97%. Model ini memiliki kinerja yang jauh lebih baik jika dibandingkan dengan model berbasis Multinomial Logistic Regression (MLR) maupun Support Vector Machine (SVM). Sesuai peta jalan penelitian, model yang dihasilkan dari penelitian ini akan digunakan untuk pengembangan sistem rekomendasi yang dapat membantu calon mahasiswa baru dalam memilih program studi saat proses penerimaan mahasiswa baru (PMB), khususnya di lingkungan UII. AbstractChoosing a major for the prospective undergraduate students is one of the most important moments in their life, especially for the high school graduates. Not infrequently, a wrong choice can lead to academic failure or even other difficulties after graduating from college. While a machine learning-based college major recommendation system is not strange in some developed countries, it is not the case in Indonesia. This study aims to serve as a pilot project for such a recommendation system by using personal and academic data of all students and alumni of the undergraduate programs in Universitas Islam Indonesia (UII) where academic achievement data is used as the ground truth label. Out of three models used and evaluated in this study, we found that Random Forest-based model to be the best option with an accuracy of 86%, precision on 84%, recall of 86%, and AUC of 97%. We also found that this model has a much better performance than other models with Multinomial Logistic Regression (MLR) or Support Vector Machine (SVM). The resulting model from this study will be deployed to develop a college major recommendation system that can help the prospective students choose their majors during college admission process, particularly in the context of UII as per research roadmap. ]]>
