| dc.contributor.author | Zain Fadillah, Riestiya | |
| dc.date.accessioned | 2020-02-18T05:11:30Z | |
| dc.date.available | 2020-02-18T05:11:30Z | |
| dc.date.issued | 2020-02-18 | |
| dc.identifier.citation | APA 6th | en_US |
| dc.identifier.uri | https://library.universitaspertamina.ac.id//xmlui/handle/123456789/1007 | |
| dc.description | This research proposed the development of the Indonesian Sign Language (Bisindo) translatormodel using machine learning technology especially the Convolutional Neural Network (CNN).In contrast to existing research, Bisindo is a sign language that is relatively wide-used by theDeaf but it is not the official sign language, so the required dataset is almost non-exist. Thepurpose of this study is to improve the accessibility of the Deaf by increasing the number ofBisindo electronic translators. Besides, this study also aimed to evaluate the parameter-transfermethod to overcome the limitations on the amount of data in the dataset and its performance inthe implementation of different sign language systems. The development of the Bisindo modelis done by utilizing the architecture of the American Sign Language (ASL) Model through twoapproaches based on the usage of the model parameters, called Model A and Model B. ModelA developed without using parameters from the ASL Model (without parameter-transfer),while the Model B developed using the ASL Model parameters and the knowledge parameterswithin (with transfer parameters). The method used in the development is experiment anddata analysis to determine the best model was done quantitatively by analyzing the duration ofthe training, accuracy during testing and F1 score at testing. The ASL model in the study hada testing accuracy of 96.40%. The results showed that Model A produced a testing accuracy of94.38% while Model B produced only 30% of testing accuracy. It can be concluded that theparameter-transfer in Model B requires less training time than Model A and makes the model beable to learn the Bisindo features that have similarities to the ASL features. However, to studythe entire Bisindo alphabet, using architecture without parameter-transfer is a better approachthan using parameter-transfer. | en_US |
| dc.description.abstract | Penelitian ini mengusulkan pengembangan model penerjemah Bahasa Isyarat Indonesia (Bi-sindo) dengan memanfaatkan teknologimachine learningkhususnyaConvolutional Neural Net-work(CNN). Berbeda dengan penelitian yang sudah ada, Bisindo adalah bahasa isyarat yangrelatif banyak digunakan Tuli namun tidak resmi, sehingga dataset yang diperlukan hampirtidak ada. Tujuan dari penelitian ini yaitu bertambahnya jumlah penerjemah Bisindo elek-tronik untuk meningkatkan aksesibilitas Tuli. Selain itu, penelitian ini juga bertujuan un-tuk mengevaluasi metodeparameter-transferuntuk mengatasi keterbatasan jumlah data padadataset dan implementasi pada sistem bahasa isyarat yang berbeda. Pengembangan model Bi-sindo dilakukan dengan memanfaatkan arsitektur Model American Sign Language (ASL) mela-lui dua pendekatan berdasarkan pemanfaatan parameter model tersebut, yang kemudian disebutModel A dan Model B. Model A dikembangkan tanpa menggunakan parameter dari Model ASL(tanpaparameter-transfer), sedangkan Model B dikembangkan dengan menggunakan parame-ter Model ASL sertaknowledge parameterdi dalamnya (denganparameter-transfer). Metodeyang digunakan dalam pengembangan adalah eksperimen dan analisis data untuk menentukanmodel terbaik secara kuantitatif dengan analisis variabel durasi training,akurasi saattestingdanF1 Scoresaattesting. Model ASL di penelitian memiliki akurasitestingsebesar 96.40%.Hasil penelitian menunjukkan bahwa Model A menghasilkan akurasitestingsebesar 94.38%sedangkan Model B sebesar 30%. Dapat disimpulkan bahwaparameter-transferpada ModelB memerlukanwaktu trainingyang lebih sedikit dibandingkan Model A serta mampu mem-pelajari fitur Bisindo yang memiliki kemiripan dengan fitur ASL. Namun, untuk mempelajarikeseluruhan alfabet Bisindo, pemanfaatan arsitektur tanpaparameter-transfermerupakan pen-dekatan yang lebih baik dibandingkan menggunakanparameter-transfer. | en_US |
| dc.language.iso | other | en_US |
| dc.subject | Tuli | en_US |
| dc.subject | Bahasa isyarat | en_US |
| dc.subject | Bisindo | en_US |
| dc.subject | Machine learning | en_US |
| dc.subject | ASL | en_US |
| dc.subject | Transfer learning | en_US |
| dc.subject | Parameter transfer | en_US |
| dc.title | Model Penerjemah Bahasa Isyarat Indonesia (Bisindo) menggunakan Convolutional Neural Network | en_US |
| dc.type | Thesis | en_US |
