IMPLEMENTASI PROTOTIPE SISTEM PENDETEKSI KEBAKARAN DAN PEMBERITAHUAN DINI TERINTEGRASI INTERNET OF THINGS DENGAN MACHINE LEARNING

ILHAM DWI SAPUTRA, . (2026) IMPLEMENTASI PROTOTIPE SISTEM PENDETEKSI KEBAKARAN DAN PEMBERITAHUAN DINI TERINTEGRASI INTERNET OF THINGS DENGAN MACHINE LEARNING. Sarjana thesis, UNIVERSITAS NEGERI JAKARTA.

	Text COVER.pdf Download (2MB)
	Text BAB I.pdf Download (348kB)
	Text BAB II.pdf Restricted to Registered users only Download (933kB) | Request a copy
	Text BAB III.pdf Restricted to Registered users only Download (846kB) | Request a copy
	Text BAB IV.pdf Restricted to Registered users only Download (913kB) | Request a copy
	Text BAB V.pdf Restricted to Registered users only Download (331kB) | Request a copy
	Text DAFTAR PUSTAKA.pdf Download (304kB)
	Text LAMPIRAN.pdf Restricted to Registered users only Download (848kB) | Request a copy

Abstract

Kebakaran rumah tangga, khususnya di area dapur, masih menjadi permasalahan keselamatan yang signifikan, dengan mayoritas kejadian dipicu oleh kelalaian saat memasak, kebocoran gas, dan peningkatan suhu yang tidak terdeteksi sejak dini. Sistem deteksi konvensional berbasis sensor tunggal dan ambang batas (threshold) tetap terbukti rentan menghasilkan alarm palsu karena tidak mampu membedakan aktivitas memasak normal dari kondisi berbahaya. Penelitian ini merancang dan mengimplementasikan prototipe sistem pendeteksi kebakaran dan pemberitahuan dini berbasis mikrokontroler ESP32 yang mengintegrasikan sensor thermal array AMG8833 dan sensor gas/asap MQ-135, diolah menggunakan algoritma machine learning Support Vector Machine (SVM), serta terhubung dengan sistem notifikasi berbasis Internet of Things (IoT) dan dashboard web React Vite. Pengujian perangkat keras menunjukkan sensor MQ-135 memiliki rata-rata akurasi 96,48% (error 3,52%) dan sensor AMG8833 memiliki rata-rata akurasi 96,54% (error 3,46%) terhadap alat ukur pembanding. Model SVM yang dilatih menggunakan 1.501 baris data dan 12 fitur turunan dari data sensor mencapai akurasi 98,67% pada data uji, dengan precision dan recall di atas 0,97 pada seluruh kelas (NORMAL, WARNING, DANGER), tanpa kesalahan klasifikasi antara kondisi NORMAL dan DANGER. Sistem mitigasi awal (buzzer dan relay) serta notifikasi IoT memiliki waktu respons rata-rata di bawah 1,2 detik pada seluruh skenario pengujian integrasi. Hasil ini menunjukkan bahwa integrasi multi-sensor dengan SVM berpotensi meningkatkan akurasi deteksi dini kebakaran dibandingkan sistem konvensional, meskipun validasi lanjutan pada kondisi dapur nyata dan dataset yang lebih bervariasi masih diperlukan sebelum kesimpulan ini dapat digeneralisasi di luar kondisi simulasi terkontrol yang digunakan pada penelitian ini.***** Household fires, particularly those originating in the kitchen, remain a significant safety concern, with most incidents triggered by unattended cooking, gas leaks, and undetected temperature increases. Conventional single-sensor, threshold-based detection systems remain prone to false alarms because they cannot distinguish normal cooking activity from genuinely hazardous conditions. This research designed and implemented a prototype early fire detection and notification system based on the ESP32 microcontroller, integrating an AMG8833 thermal array sensor and an MQ-135 gas/smoke sensor, processed using a Support Vector Machine (SVM) classification algorithm, and connected to an Internet of Things (IoT) notification system with a React Vite web dashboard. Hardware testing showed the MQ-135 sensor achieved an average accuracy of 96.48% (3.52% error) and the AMG8833 sensor an average accuracy of 96.54% (3.46% error) against reference measurement instruments. The SVM model, trained on 1,501 rows of data using 12 derived sensor features, achieved 98.67% accuracy on the test set, with precision and recall above 0.97 across all three classes (NORMAL, WARNING, DANGER), with no misclassifications between NORMAL and DANGER conditions. The early mitigation system (buzzer and relay) and IoT notification achieved an average response time under 1.2 seconds across all integration test scenarios. These results indicate that multi-sensor integration with SVM has the potential to improve early fire detection accuracy compared to conventional systems, although further validation under real kitchen conditions and with more diverse datasets is still required before these findings can be generalized beyond the controlled simulation conditions used in this study

Item Type:	Thesis (Sarjana)
Additional Information:	1). Syufrijal, S.T., M.T. 2). Dr. Widodo, S.Kom, M.Kom
Subjects:	Teknologi dan Ilmu Terapan > Teknik Elektronika Teknologi dan Ilmu Terapan > Teknik Komputer
Divisions:	FT > D IV Teknologi Rekayasa Otomasi
Depositing User:	Ilham Dwi Saputra .
Date Deposited:	06 Aug 2026 07:41
Last Modified:	06 Aug 2026 07:41
URI:	http://repository.unj.ac.id/id/eprint/68802

Actions (login required)

View Item