Comparing Outlier Detection Methods: An Application on Indonesian Air Quality Data

Anwar Fitrianto (Department of Statistics, IPB University)
Amalia Kholifatunnisa (Department of Statistics, IPB University, Bogor, West Java)
Anang Kurnia (2Institute of Engineering Mathematics, Universiti Malaysia Perlis)

CAUCHY: Jurnal Matematika Murni dan Aplikasi
Vol 9, No 2 (2024): CAUCHY: JURNAL MATEMATIKA MURNI DAN APLIKASI

There are many methods for detecting outliers, but only a few methods consider data distribution. This research compares outlier detection method on univariate data with a skewed distribution. Outlier detection methods used in this research are Tukey's boxplot, adjusted boxplot, sequential fences, and adjusted sequential fences. It identifies areas of concern due to poor air quality during the Implementation of Micro-Community Activity Restrictions. The study used Indonesian air quality index data.The adjusted boxplot method performs best based on the number of outliers detected, error rate, accuracy, precision, specificity, sensitivity, and robustness. Adjusted boxplot and adjusted sequential fences can detect tails that contain outliers accurately because the skewness coefficient makes them more robust. Meanwhile, Tukey's boxplot and sequential fences are poor methods since they couldn’t detect correctly true outliers. Based on the results, adjusted boxplot is the best method. Then, areas that need attention due to poor air quality include South Sumatera, South Sulawesi, West Java, Riau, North Sumatera, Jambi, Jakarta, and East Java.