AN AI-BASED DECISION SUPPORT SYSTEM FOR PRIORITIZING EMERGENCY INFORMATION DURING DISASTERS USING MULTI-LABEL CLASSIFICATION AND EXPLAINABLE PRIORITY RULES

During large-scale natural disasters and humanitarian crises, emergency response agencies, 911 dispatch coordinators, and disaster relief non-governmental organizations (NGOs) are inundated with thousands of unstructured incident reports and survivor pleas. Manually reviewing incoming messages leads to severe cognitive fatigue, operational dispatch delays, and triage paralysis. Existing automated frameworks often treat disaster multi-label classification as an isolated academic exercise without actionable priority ranking, force supervised machine learning onto unannotated priority labels, or make dangerous claims of autonomous emergency dispatch. This paper presents an end-to-end, academic decision-support framework and interactive responder operations system designed to ingest, clean, categorize, prioritize, and verify disaster-response incident messages in real time. The proposed system benchmarks three supervised multi-label learning paradigms—Multi-Output Logistic Regression, Linear Support Vector Machines (LinearSVC), and Deep Learning Multi-Layer Perceptrons (MLP)—across 34 active emergency categories on a benchmark dataset of 26,242 disaster messages. Based on validation Micro F1 scores (0.6400), the optimal neural architecture is selected and evaluated on an unseen test set, achieving a Micro F1 of 0.6912 and a Micro Precision of 0.8059. To bridge classification outputs with actionable triage, we introduce a transparent, auditable Proposed Emergency Priority Framework that computes numerical priority scores (0–100) and assigns discrete priority tiers (CRITICAL, HIGH, MEDIUM, LOW) using domain severity weights, operational modifiers, and regex distress keyword signals. Crucially, the system enforces a strict Human-in-the-Loop (HITL) protocol where predictions remain in an 'Awaiting Verification' state until validated by authorized dispatch personnel through a full-stack Flask/SQLite web operations dashboard.

Authors

Publication Details

Journal
EPRA International Journal of Research & Development (IJRD)
Published
2026-11-17
DOI
https://doi.org/10.36713/epra32143
Primary Topic
Text and Document Classification Technologies
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

AN AI-BASED DECISION SUPPORT SYSTEM FOR PRIORITIZING EMERGENCY INFORMATION DURING DISASTERS USING MULTI-LABEL CLASSIFICATION AND EXPLAINABLE PRIORITY RULES

Prof. Dr. Mrs. Shivani A. Budhkar, Triveni Jadhav, Sumit Aswar, Anuja Dokh et al.
2 links
EPRA International Journal of Research & Development (IJRD)
Text and Document Classification Technologies
article

AN AI-BASED DECISION SUPPORT SYSTEM FOR PRIORITIZING EMERGENCY INFORMATION DURING DISASTERS USING MULTI-LABEL CLASSIFICATION AND EXPLAINABLE PRIORITY RULES

Prof. Dr. Mrs. Shivani A. Budhkar, Triveni Jadhav, Sumit Aswar, Anuja Dokh, Sakshi Gate
article en

Abstract

During large-scale natural disasters and humanitarian crises, emergency response agencies, 911 dispatch coordinators, and disaster relief non-governmental organizations (NGOs) are inundated with thousands of unstructured incident reports and survivor pleas. Manually reviewing incoming messages leads to severe cognitive fatigue, operational dispatch delays, and triage paralysis. Existing automated frameworks often treat disaster multi-label classification as an isolated academic exercise without actionable priority ranking, force supervised machine learning onto unannotated priority labels, or make dangerous claims of autonomous emergency dispatch. This paper presents an end-to-end, academic decision-support framework and interactive responder operations system designed to ingest, clean, categorize, prioritize, and verify disaster-response incident messages in real time. The proposed system benchmarks three supervised multi-label learning paradigms—Multi-Output Logistic Regression, Linear Support Vector Machines (LinearSVC), and Deep Learning Multi-Layer Perceptrons (MLP)—across 34 active emergency categories on a benchmark dataset of 26,242 disaster messages. Based on validation Micro F1 scores (0.6400), the optimal neural architecture is selected and evaluated on an unseen test set, achieving a Micro F1 of 0.6912 and a Micro Precision of 0.8059. To bridge classification outputs with actionable triage, we introduce a transparent, auditable Proposed Emergency Priority Framework that computes numerical priority scores (0–100) and assigns discrete priority tiers (CRITICAL, HIGH, MEDIUM, LOW) using domain severity weights, operational modifiers, and regex distress keyword signals. Crucially, the system enforces a strict Human-in-the-Loop (HITL) protocol where predictions remain in an 'Awaiting Verification' state until validated by authorized dispatch personnel through a full-stack Flask/SQLite web operations dashboard.

EPRA International Journal of Research & Development (IJRD)
Text and Document Classification Technologies
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon