KLCBL: An Improved Police Incident Classification Model

TL;DR

This paper introduces KLCBL, a multichannel neural network that combines advanced text preprocessing and deep learning techniques to improve police incident classification accuracy, addressing manual inefficiencies and system limitations.

Contribution

The paper presents a novel multichannel neural network model, KLCBL, integrating KAN, LERT, CNN, and BiLSTM for enhanced police incident classification.

Findings

Achieved 91.9% accuracy on real data

Outperformed baseline models in classification tasks

Addresses challenges in police data processing

Abstract

Police incident data is crucial for public security intelligence, yet grassroots agencies struggle with efficient classification due to manual inefficiency and automated system limitations, especially in telecom and online fraud cases. This research proposes a multichannel neural network model, KLCBL, integrating Kolmogorov-Arnold Networks (KAN), a linguistically enhanced text preprocessing approach (LERT), Convolutional Neural Network (CNN), and Bidirectional Long Short-Term Memory (BiLSTM) for police incident classification. Evaluated with real data, KLCBL achieved 91.9% accuracy, outperforming baseline models. The model addresses classification challenges, enhances police informatization, improves resource allocation, and offers broad applicability to other classification tasks.