|
Real-Time Classification Improvement ofIndonesian Sign System Letters (SIBI)Using K-Nearest Neighbor Algorithm |
|---|---|
| รหัสดีโอไอ | |
| Creator | Oktaf Agni Dhewa |
| Title | Real-Time Classification Improvement ofIndonesian Sign System Letters (SIBI)Using K-Nearest Neighbor Algorithm |
| Contributor | Safitri Yuliana Utama, Aris Nasuha,Teddy Surya Gunawan, Gilang Nugraha Putu Pratama |
| Publisher | Research Administration Division |
| Publication Year | 2567 |
| Journal Title | Science & Technology Asia |
| Journal Vol. | 29 |
| Journal No. | 3 |
| Page no. | 94-114 |
| Keyword | Distance Metric, Lazy learner system, Machine learning, Sign language |
| URL Website | https://www.tci-thaijo.org/index.php/SciTechAsia/index |
| Website title | ThaiJo2 |
| ISSN | 2586-9000 |
| Abstract | Indonesian Sign Language (SIBI) is a vital means of communication for individualswith hearing impairments. The automatic translation from spoken language to SIBI presentschallenges in accurately predicting sign characters. The information transfer process be-comes biased when system predictions are incorrect. The current approach lacks accuracydue to data variations that may lead to character similarities. This research addresses thisissue with an improved method incorporating linguistic features and contextual information.A novel approach is introduced to enhance SIBI character predictions using the K-NearestNeighbor (K-NN) algorithm. The K-NN algorithm is employed to predict the most suit-able SIBI character based on the similarity of linguistic features between input speech andexisting data. This research compares distance metrics such as Euclidean, Manhattan, andChebyshev to determine the optimal number of nearest neighbors (K) for achieving the mostaccurate outcomes. Experimental results employing 200 data points per label yielded satis-factory average predictions for each label. The experiments underscore the effectiveness ofthe K-NN model utilizing the Chebyshev distance metric with K=7 on the 200 data labels,as it provided excellent probability results for each label. |
