Implementation of Split Sampling for Decision Tree and K-Nearest Neighbor Algorithms in the DKI Jakarta Legislative Election

Abstrak

The 2009 legislative election was contested by 44 political parties, consisting of national and local parties. In the 2009 Legislative Election for DKI Jakarta, there were 2,268 candidates for the Regional House of Representatives (DPRD) from 44 parties competing for 94 seats in the DKI Jakarta Regional People's Representative Council. Data mining is a series of processes aimed at discovering added value in the form of information that has not been previously known manually from a database. The classification method can be used to predict the results of legislative elections. In this study, the author employs the Decision Tree and K-Nearest Neighbor classification algorithms. This research utilizes several data sampling techniques, namely Linear Sampling, Shuffled Sampling, and Stratified Sampling. The data split partition used in this study was 80% for training and 20% for testing. The software tool utilized was RapidMiner. The performance variables measured include Recall, Precision, and Accuracy. The results of this study indicate that, overall, Linear Split Sampling outperforms Shuffled Split Sampling and Stratified Split Sampling. For the Decision Tree algorithm, Linear Split Sampling achieved a Recall of 100%, Precision of 82.05%, and Accuracy of 98.46%. Shuffled Split Sampling recorded a Recall of 81.82%, Precision of 85.71%, and Accuracy of 98.46%. Meanwhile, Stratified Split Sampling obtained a Recall of 100%, Precision of 82.05%, and Accuracy of 97.80%. Meanwhile, for the K-Nearest Neighbor (KNN) algorithm, Linear Split Sampling achieved a Recall of 93.75%, Precision of 75%, and Accuracy of 97.36%. Shuffled Split Sampling recorded a Recall of 59.09%, Precision of 81.25%, and Accuracy of 97.36%. Stratified Split Sampling obtained a Recall of 31.58%, Precision of 85.71%, and Accuracy of 96.92%.