Klasifikasi Panggilan Darurat 112 Kota Mojokerto Menggunakan Analisis Transkrip ASR Dan Metode TF-IDF - Linear SVM

Sampurno, Suryo Hadi (2026) Klasifikasi Panggilan Darurat 112 Kota Mojokerto Menggunakan Analisis Transkrip ASR Dan Metode TF-IDF - Linear SVM. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6022241094-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (6MB) | Request a copy

Abstract

Layanan Nomor Tunggal Panggilan Darurat 112 berperan penting dalam penanganan kejadian gawat darurat, namun meningkatnya volume panggilan serta karakteristik percakapan yang tidak terstruktur menjadikan proses identifikasi kategori kejadian semakin kompleks. Kondisi ini menimbulkan kebutuhan akan pendekatan berbasis data yang mampu menganalisis transkrip panggilan secara konsisten. Penelitian ini mengembangkan dan mengevaluasi pipeline klasifikasi berbasis teks menggunakan transkrip Automatic Speech Recognition (ASR) dari panggilan 112 Kota Mojokerto. Data penelitian mencakup 807 panggilan darurat periode Desember 2022-Agustus 2025 yang diklasifikasikan ke dalam 12 kategori kejadian. Proses ASR menggunakan layanan Google Web Speech API menghasilkan rata-rata Word Error Rate (WER) sebesar 0,4730, dengan segmentasi 15 detik memberikan kualitas transkripsi terbaik (WER 0,4280). Transkrip kemudian diproses melalui normalisasi struktural dan normalisasi leksikal sehingga mengurangi elemen noise serta menyeragamkan variasi kata. Data dibagi secara terstratifikasi (70:30), diekstraksi menggunakan Term Frequency-Inverse Document Frequency (TF-IDF) dengan 3.779 fitur, dan ditangani menggunakan cost-sensitive learning serta SMOTE untuk mengatasi ketidakseimbangan kelas. Dua algoritma klasifikasi diuji, yaitu Linear Support Vector Machine (SVM) dan Logistic Regression. Hasil menunjukkan bahwa SVM dengan SMOTE merupakan konfigurasi terbaik dengan akurasi 0,85 dan F1-score 0,81, sedangkan Logistic Regression dengan cost-sensitive learning menghasilkan akurasi 0,83 dan F1-score 0,75. Model SVM juga menunjukkan performa yang lebih stabil pada kategori minoritas. Penelitian ini menunjukkan bahwa kombinasi ASR, TF-IDF, dan SVM dapat membentuk sistem klasifikasi multi-kelas yang efektif meskipun data memiliki ketidakseimbangan kategori dan tingkat kesalahan transkripsi yang tinggi. Temuan ini memberikan dasar awal bagi pengembangan sistem pendukung analisis panggilan darurat di Indonesia.
============================================================================================================================
Emergency Call Number 112 plays an essential role in handling lifethreatening incidents, yet the increasing volume of calls and the unstructured nature of conversations make the identification of emergency categories increasingly complex. This situation highlights the need for a data-driven approach capable of consistently analyzing call transcripts. This study develops and evaluates a text-based classification pipeline using Automatic Speech Recognition (ASR) transcripts from 112 calls in Mojokerto City. The dataset consists of 807 emergency calls recorded between December 2022 and August 2025, categorized into 12 emergency types. The ASR process, using the Google Web Speech API, achieved an average Word Error Rate (WER) of 0.4730, with the 15-second segmentation providing the best transcription quality (WER 0.4280). The transcripts were then processed through structural and lexical normalization to reduce noise elements and standardize word variations. The data were split using a stratified 70:30 ratio, transformed using Term Frequency–Inverse Document Frequency (TF-IDF) with 3,779 features, and handled with cost-sensitive learning and SMOTE to address class imbalance. Two classification algorithms were evaluated, namely Linear Support Vector Machine (SVM) and Logistic Regression. The results show that SVM combined with SMOTE yielded the best performance with an accuracy of 0.85 and an F1-score of 0.81, while Logistic Regression with cost-sensitive learning achieved an accuracy of 0.83 and an F1-score of 0.75. SVM also demonstrated more stable performance across minority classes. This study demonstrates that the combination of ASR, TF-IDF, and SVM can form an effective multi-class classification system despite high class imbalance and transcription errors. These findings provide an initial foundation for developing analytical decision-support tools for emergency call services in Indonesia.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Panggilan darurat 112, Automatic Speech Recognition, TF-IDF, Support Vector Machine, SMOTE, Emergency call 112, Automatic Speech Recognition, TF-IDF, Support Vector Machine, SMOTE
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20101-(S2) Master Thesis
Depositing User:	Suryo Hadi Sampurno
Date Deposited:	19 Jan 2026 05:50
Last Modified:	19 Jan 2026 05:50
URI:	http://repository.its.ac.id/id/eprint/129709

Actions (login required)

View Item