Electronic Theses and Dissertation

Kehilangan data (missing values) merupakan permasalahan umum dalam analisis data deret waktu musiman yang dapat menurunkan akurasi model dan mengganggu interpretasi pola musiman. Penelitian ini bertujuan untuk menganalisis dan membandingkan performa dua metode khusus untuk imputasi data hilang pada data deret waktu musiman, yaitu Seasonally Decomposed Missing Value Imputation (SD-MVI) dan Seasonally Splitted Missing Value Imputation (SS-MVI). Tiga dataset musiman digunakan dalam studi ini: data harian temperatur permukaan di Banda Aceh, data bulanan jumlah penumpang pesawat di AS, dan data bulanan suhu El Nino di Pasifik Tengah Ekuator. Data hilang disimulasikan secara acak dalam proporsi 5% hingga 25%. Evaluasi kinerja metode dilakukan menggunakan tiga metrik evaluasi: Root Mean Square Error (RMSE), Weighted Absolute Percentage Error (WAPE), dan Mean Absolute Error (MAE). Hasil penelitian menunjukkan bahwa metode SD-MVI secara konsisten memberikan hasil yang lebih akurat dibandingkan SS-MVI pada semua proporsi data hilang dan seluruh dataset. SD-MVI terbukti lebih mampu mempertahankan distribusi dan variasi data asli, terutama pada proporsi kehilangan data yang tinggi. Temuan ini memberikan rekomendasi bahwa SD-MVI merupakan metode yang lebih efektif untuk mengatasi data hilang dalam deret waktu musiman, khususnya dalam konteks analisis yang memerlukan ketelitian terhadap pola musiman.

Kata kunci: Imputasi, Data hilang, SD-MVI, SS-MVI, Musiman

Missing values are a common problem in the analysis of seasonal time series data that can reduce model accuracy and interfere with the interpretation of seasonal patterns. This study aims to analyze and compare the performance of two specific methods for missing data imputation on seasonal time series data, namely Seasonally Decomposed Missing Value Imputation (SD-MVI) and Seasonally Splitted Missing Value Imputation (SS-MVI). Three seasonal datasets are used in this study: daily surface temperature data in Banda Aceh, monthly airplane passenger count data in the US, and monthly El Nino temperature data in the Equatorial Central Pacific. Missing data was simulated randomly in the proportion of 5% to 25%. The performance evaluation of the method was conducted using three evaluation metrics: Root Mean Square Error (RMSE), Weighted Absolute Percentage Error (WAPE), and Mean Absolute Error (MAE). The results show that the SD-MVI method consistently provides more accurate results than SS-MVI for all proportions of missing data and all datasets. SD-MVI proved to be better able to preserve the distribution and variation of the original data, especially at a high proportion of missing data. The findings suggest that SD-MVI is a more effective method for dealing with missing data in seasonal time series, especially in the context of analyses that require attention to seasonal patterns. Keywords: Imputation, Missing values, SD-MVI, SS-MVI, Seasonal