E-government Public Complaints Text Classification Using Particle Swarm Optimization in Naive Bayes Algorithm

Abstract

A web-based online complaint portal is one of the e-government public services. The complaint's content must be categorised in order for it to be transmitted to the appropriate agency swiftly and properly. The most often used standard classification algorithms are the Naive Bayes Classifier (NBC) and k-Nearest Neighbor (k-NN), both of which classify just one label and must be tuned. The purpose of this project is to categorize complaint messages that include several labels simultaneously using NBC tuned for particle swarm optimization (PSO). The data source is the Open Data Jakarta and is partitioned into 70% training data and 30% test data for classification into seven labels. The NBC and k-NN algorithms are used to compare PSO's optimization performance. Cross-validation ten times revealed that optimizing NBC with PSO obtained an accuracy of 88.16%, much superior than k-NN at 83% and NBC at 70.57%. This optimization approach may be used to improve the efficacy of community-based e-government services.