Klasifikasi Emosi Manusia dengan Metode Ensemble Stacking pada Data Transkrip Audio-ke-teks

Subhan, Muhammad (2023) Klasifikasi Emosi Manusia dengan Metode Ensemble Stacking pada Data Transkrip Audio-ke-teks. Other thesis, Institut Teknologi Sepuluh November.

Text 05111940000204-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 September 2025. Download (4MB) | Request a copy

Abstract

Pemrosesan Bahasa Alami (Natural Language Processing/NLP) telah menjadi topik yang menarik dalam bidang kecerdasan buatan dan analisis data. Salah satu tugas penting dalam NLP adalah mendeteksi emosi manusia berdasarkan data teks. Dalam penelitian ini, kami menggunakan beberapa model tunggal seperti SVM, Random Forest, dan MLP, dengan berbagai metode ekstraksi fitur seperti Bag-of-Words (BoW), Word2Vec, dan BERT. Selain itu, kami juga menerapkan ensemble method menggunakan Voting Classifier dan Stacking Classifier untuk meningkatkan performa prediksi. Pada penelitian ini menekankan pentingnya penggunaan ensemble method dalam meningkatkan performa dalam mendeteksi emosi menggunakan data audio. Hasil evaluasi menunjukkan bahwa ensemble method mencapai akurasi yang lebih tinggi dibandingkan dengan model tunggal. Hasil tersebut didapatkan melalui percobaan pada data Audio Speech Sentiment yang telah diubah menjadi teks menggunakan Speech Recognition. Penggabungan model menjadi ensemble method dapat menghasilkan prediksi yang lebih akurat dalam mendeteksi emosi dalam data teks. Hasil evaluasi menunjukkan bahwa pada model tunggal menggunakan data teks, (Random Forest dengan fitur Bag-of-Words (BoW) mencapai akurasi F1-Score tertinggi sebesar 85%. Namun, pada ensemble method dengan teknik stacking, kami mencapai akurasi yang lebih tinggi sebesar F1-Score 87%, menjadikannya sebagai model terbaik di antara semua model yang dievaluasi. Pada penelitian dihindari penggunaan model SVM-BOW dalam ensemble method karena memiliki keterbatasan dalam adaptasi dan kompleksitasnya. Penggunaan F1-Score sebagai metrik evaluasi yang dipilih menunjukkan bahwa penelitian ini mempertimbangkan keseimbangan antara presisi dan recall dalam kasus klasifikasi multiclass
==================================================================================================================================
Natural Language Processing (NLP) has become an intriguing topic in the field of artificial intelligence and data analysis. One important task in NLP is detection of human emotions based on textual data. In this research, we employed several single models such as SVM, Random Forest, and MLP, along with various feature extraction methods including Bagof-Words (BoW), Word2Vec, and BERT. Additionally, we also applied ensemble methods using Voting Classifier and Stacking Classifier to enhance prediction performance. The evaluation results showed that among the single models using textual data, Random Forest with BoW features achieved the highest F1-Score accuracy of 85%. However, with ensemble methods using stacking technique, we achieved even higher accuracy with an F1- Score of 87%, making it the best-performing model among all evaluated models. In this research, the use of SVM-BoW model in the ensemble method was avoided due to its limitations in adaptation and complexity. This research emphasizes the importance of employing ensemble methods to enhance the performance of emotion detection using audio data. The evaluation results demonstrate that ensemble methods achieve higher accuracy compared to single models. These results were obtained through experiments on Audio Speech Sentiment data that were transformed into text using Speech Recognition. By combining predictions from multiple models or algorithms, the model can generate more accurate predictions in detecting emotions within textual data. The use of F1-Score as the chosen evaluation metric indicates that this study considered the balance between precision and recall in the case of multiclass classification

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Ekstraksi fitur, Ensemble method, Model tunggal, Natural Language Processing.Ensemble method, Feature Extraction, Natural Language Processing, Single model.
Subjects:	T Technology > T Technology (General) > T174 Technological forecasting T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Muhammad Subhan
Date Deposited:	04 Aug 2023 03:01
Last Modified:	04 Aug 2023 03:01
URI:	http://repository.its.ac.id/id/eprint/100546

Actions (login required)

View Item