Pengenalan Fonik Dengan Metode Berbasis Tree

Firoos, Meyroja Jovancha (2025) Pengenalan Fonik Dengan Metode Berbasis Tree. Other thesis, Insititut Teknologi Sepuluh Nopember.

Text 5025211204-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Penelitian ini bertujuan untuk mengembangkan dan mengevaluasi sistem pengenalan kata fonik menggunakan pendekatan machine learning berbasis pohon. Sistem dirancang untuk mengenali 17 kelas ejaan kata dalam bahasa Inggris, seperti cat, dog, dan sun, berdasarkan sinyal suara yang telah diekstraksi fitur foniknya. Dua metode ekstraksi fitur yang digunakan adalah Mel-Frequency Cepstral Coefficients (MFCC) dan Perceptual Linear Prediction (PLP), yang masing-masing menghasilkan representasi vektor sepanjang 7.020 elemen dari matriks fitur 39×180. Dataset dikumpulkan melalui kombinasi fonem terbatas menggunakan parameter PRODUCT_LIMIT = 1500, lalu diperluas melalui augmentasi pola suara seperti noise, time shifting, pitch shifting, dan gain adjustment, sehingga total sampel mencapai 76.500 data suara. Model yang digunakan dalam klasifikasi adalah Decision Tree, Random Forest, dan XGBoost, yang dievaluasi menggunakan teknik Stratified K-Fold Cross Validation serta metrik akurasi dan F1-score. Hasil evaluasi menunjukkan bahwa model XGBoost dengan fitur MFCC memberikan akurasi tertinggi sebesar 99,24%, diikuti oleh Random Forest sebesar 98,29%, dan Decision Tree sebesar 94,04%. Lalu, model XGBoost dengan PLP mendapat akurasi sebesar 98,91%, diikuti Random Forest sebesar 98,40% dan Decision Tree sebesar 94,73%. Analisis inferensi pada data uji menunjukkan bahwa sistem mampu mengklasifikasikan kata dengan tingkat keyakinan tinggi, bahkan pada kelas dengan F1-score pelatihan terendah, seperti kata cat. Dengan hasil tersebut, sistem yang dibuat terbukti efektif untuk tugas klasifikasi fonik berbasis kata menggunakan model berbasis pohon, serta memiliki potensi untuk diterapkan dalam aplikasi pembelajaran fonik interaktif dan edukatif dengan efisiensi tinggi.
========================================================================================================================
This study aims to develop and evaluate a phonics word recognition system using a tree- based machine learning approach. The system is designed to recognize 17 simple English word classes, such as cat, dog, and sun, based on audio signals whose phonetic features have been extracted. Two feature extraction methods were used: Mel-Frequency Cepstral Coefficients (MFCC) and Perceptual Linear Prediction (PLP), each producing a vector representation of 7,020 elements from a 39×180 feature matrix. The dataset was constructed through limited phoneme combinations using a parameter of PRODUCT_LIMIT = 1500, and expanded through sound pattern augmentation techniques such as noise, time shifting, pitch shifting, and gain adjustment, resulting in a total of 76,500 voice samples. The classification models used were Decision Tree, Random Forest, and XGBoost, evaluated using Stratified K-Fold Cross Validation and performance metrics including accuracy and F1-score. The evaluation results showed that the XGBoost model with MFCC features achieved the highest accuracy of 99.24%, followed by Random Forest with MFCC at 98.29%, and Decision Tree with MFCC at 94.04%. Meanwhile, XGBoost with PLP achieved an accuracy of 98.91%, followed by Random Forest with PLP at 98.40%, and Decision Tree with PLP at 94.73%. Inference analysis on the test data revealed that the system was capable of classifying words with high confidence, even in classes with the lowest training F1-scores, such as the word cat. These results confirm that the developed system is effective for word-level phonics classification using tree-based models and has the potential to be applied in interactive and educational phonics learning applications with high efficiency.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Pengenalan fonik, Machine learning, Decision Tree, Random Forest, XGBoost, MFCC, PLP, Ekstraksi fitur audio, Augmentasi data suara, Klasifikasi fonetik, Phonics recognition, Machine learning, Decision Tree, Random Forest, XGBoost, MFCC, PLP, Audio feature extraction, Audio data augmentation, Phonetic classification
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Meyroja Jovancha Firoos
Date Deposited:	31 Jul 2025 05:40
Last Modified:	31 Jul 2025 05:40
URI:	http://repository.its.ac.id/id/eprint/124694

Actions (login required)

View Item