Detection of Violations on Public Reporting to Police with CNN-BiLSTM Hybrid Architecture Development

Abstract

The Jambi Regional Police Department (Polda Jambi) has introduced an innovative system for managing public complaints via WhatsApp. However, this implementation faces challenges due to the manual documentation process, leading to poorly documented and inaccurate complaints. To address these issues, this study proposes the development of a text mining-based system utilizing deep learning to facilitate the accurate categorization of public complaints, thereby streamlining the police's processing of these reports. Deep learning, specifically through Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory (BiLSTM), offers a robust framework for learning patterns from complaint data. This study develops and optimizes a CNN-BiLSTM architecture, involving the adjustment of layer configurations and the application of early stopping techniques to prevent overfitting. The proposed architecture is tested on two datasets: public complaints submitted to the Indonesian National Police via WhatsApp and Tweets from social media X. Experimental results indicate high performance across both datasets, with the architecture achieving a peak accuracy of 99% on the police data and 79% on the Twitter data. The highest-performing model is then integrated into a graphical user interface (GUI) using Streamlit, enabling the efficient and accurate classification of public complaints. This system demonstrates significant potential for enhancing the efficiency and accuracy of complaint management processes within law enforcement agencies. The findings suggest that integrating advanced deep learning techniques into public complaint systems can substantially improve the documentation and categorization of complaints, providing a scalable solution for law enforcement operations.