Evaluasi dan optimasi random forest untuk keputusan leveling pada pelatihan hospitality adaptif berbasis vr untuk anak down syndrome
Penerbit : FTI - Usakti
Kota Terbit : Jakarta
Tahun Terbit : 2026
Pembimbing 1 : Dian Pratiwi
Pembimbing 2 : Anung Barlianto Ariwibowo
Kata Kunci : Random Forest, Virtual Reality, Down Syndrome, leveling adaptif, hospitality training, probabilitas
Status Posting : Published
Status : Lengkap
| No. | Nama File | Hal. | Link |
|---|---|---|---|
| 1. | 2026_SK_STF_064002200010_Halaman-Judul.pdf | 9 | |
| 2. | 2026_SK_STF_064002200010_Surat-Pernyataan-Revisi-Terakhir.pdf | 1 | |
| 3. | 2026_SK_STF_064002200010_Surat-Hasil-Similaritas.pdf | 1 | |
| 4. | 2026_SK_STF_064002200010_Halaman-Pernyataan-Persetujuan-Publikasi-Tugas-Akhir-untuk-Kepentingan-Akademis.pdf | 1 | |
| 5. | 2026_SK_STF_064002200010_Lembar-Pengesahan.pdf | 1 | |
| 6. | 2026_SK_STF_064002200010_Pernyataan-Orisinalitas.pdf | 1 | |
| 7. | 2026_SK_STF_064002200010_Formulir-Persetujuan-Publikasi-Karya-Ilmiah.pdf | 1 | |
| 8. | 2026_SK_STF_064002200010_Bab-1-Pendahuluan.pdf | 9 | |
| 9. | 2026_SK_STF_064002200010_Bab-2-Landasan-Teori.pdf | 24 |
|
| 10. | 2026_SK_STF_064002200010_Bab-3-Metodologi-Penelitian.pdf | 34 |
|
| 11. | 2026_SK_STF_064002200010_Bab-4-Analisis-dan-Pembahasan.pdf | 24 |
|
| 12. | 2026_SK_STF_064002200010_Bab-5-Kesimpulan-dan-Saran.pdf | 2 | |
| 13. | 2026_SK_STF_064002200010_Daftar-Pustaka.pdf | 5 |
P Penelitian ini membahas evaluasi dan optimasi random forest untuk mendukung keputusan leveling adaptif pada pelatihan hospitality berbasis virtual reality (vr) bagi anak down syndrome. vr digunakan sebagai media pelatihan dan sumber data perilaku pengguna, sedangkan fokus penelitian diarahkan pada evaluasi model machine learning, bukan pada pengembangan teknis sistem vr. data yang digunakan berupa data session-level yang terdiri dari enam fitur perilaku, yaitu current_level, avg_hand_velocity, max_hand_jerk, hesitation_time, focus_consistency, dan total_duration. data periode januari hingga april sebanyak 406 sesi digunakan sebagai data pelatihan, sedangkan data potads periode juni sebanyak 123 sesi digunakan sebagai data uji eksternal. label kelulusan dibentuk menggunakan pendekatan rule-based risk score yang diinformasikan oleh observasi awal dan masukan pendamping, sehingga label diposisikan sebagai label operasional, bukan diagnosis klinis. model baseline random forest memperoleh accuracy sebesar 0,8211 dan f1-score sebesar 0,8308 pada data potads juni. setelah optimasi hyperparameter menggunakan randomizedsearchcv, model optimized memperoleh accuracy sebesar 0,8455 dan f1-score sebesar 0,8613. brier score juga menurun dari 0,1390 menjadi 0,1282, yang menunjukkan kualitas probabilitas model optimized lebih baik. hasil analisis menunjukkan bahwa fitur seperti total_duration, avg_hand_velocity, focus_consistency, hesitation_time, dan max_hand_jerk berkontribusi terhadap prediksi model. probabilitas kelas lulus digunakan sebagai dasar keputusan leveling adaptif dengan kategori naik level, naik level dengan catatan, evaluasi tambahan, dan ulang level. model hasil optimasi juga disiapkan untuk deployment ke sistem vr melalui artefak model, scaler, metadata, dan export model ke c#. hasil penelitian menunjukkan bahwa random forest dapat digunakan sebagai sistem pendukung keputusan untuk evaluasi performa pengguna pada pelatihan hospitality adaptif berbasis vr, dengan tetap menempatkan hasil model sebagai pendukung keputusan, bukan pengganti penilaian ahli.
T This study discusses the evaluation and optimization of random forest to support adaptive leveling decisions in virtual reality (vr)-based hospitality training for children with down syndrome. vr is used as a training medium and a source of user behavior data, while the research focuses on machine learning model evaluation rather than the technical development of the vr system. the dataset consists of session-level behavioral features, including current_level, avg_hand_velocity, max_hand_jerk, hesitation_time, focus_consistency, and total_duration. a total of 406 sessions collected from january to april were used as training data, while 123 potads sessions collected in june were used as external test data. the pass/fail labels were generated using a rule-based risk score informed by initial observation and caregiver input; therefore, the labels are treated as operational labels rather than clinical diagnoses. the baseline random forest model achieved an accuracy of 0.8211 and an f1-score of 0.8308 on the potads test data. after hyperparameter optimization using randomizedsearchcv, the optimized model achieved an accuracy of 0.8455 and an f1-score of 0.8613. the brier score also decreased from 0.1390 to 0.1282, indicating better probability quality. the analysis shows that features such as total_duration, avg_hand_velocity, focus_consistency, hesitation_time, and max_hand_jerk contribute to the model prediction. the probability of the pass class is used as the basis for adaptive leveling decisions, consisting of level up, level up with notes, additional evaluation, and repeat level. the optimized model is also prepared for deployment into the vr system through model artifacts, scaler, metadata, and c# model export. the results indicate that random forest can be used as a decision support system for evaluating user performance in vr-based adaptive hospitality training, while still positioning the model output as supportive information rather than a replacement for expert judgment.