A Comparative Study on Machine Learning based Prediction Models for Public Participation Rate in an Election Voting

Prediction of public participation in elections is one measure of election success. Voter participation is at the polling station level and involves four data sources: voters, polling stations, recapitulation, and village profiles. The preprocessing stage is carried out on each dataset, including maintenance, transformation, and integration. Two types of datasets are defined, involving all attributes and removing the result from attribute correlation. Classification method with five machine learning algorithms (ML) with participation prediction classes labelled High and Low. The highest result is 85.90% for the type 1 dataset and Artificial Neural Network (ANN) algorithm with 60% training and 40% testing split dataset. Furthermore, for detachment type 2, by eliminating several attributes, 100% results are obtained for the K-Nearest Neighbor (kNN) algorithm with a split dataset of 70% training and 30% testing. Of the five ML algorithms, only the Naïve Bayes (NB) algorithm did not experience an increase in prediction results. Furthermore, the significant influence of attributes on the prediction class is shown in each attribute of the dataset, including the Permanent Voter List at the Polling Place (DPT TPS), Local, Health Access and Total Recapitulation.