Electronic Theses and Dissertation

Data hilang merupakan permasalahan umum dalam data iklim yang dapat menurunkan akurasi analisis lanjutan, terutama dalam pengujian hubungan prediktif antarvariabel. Penelitian ini bertujuan mengevaluasi kinerja metode imputasi berbasis algoritma metaheuristik, yaitu Genetic Algorithm (GA), Particle Swarm Optimization (PSO), dan Differential Evolution (DE), serta menganalisis hubungan prediktif antarvariabel iklim menggunakan uji kausalitas Granger. Data yang digunakan berupa data iklim harian dari Stasiun Klimatologi Kabupaten Mempawah, Kalimantan Barat, periode 1990-2025 dengan 13.149 observasi. Variabel yang dianalisis meliputi temperatur udara, kelembapan, curah hujan, lama penyinaran matahari, dan kecepatan angin. Evaluasi dilakukan melalui simulasi penghilangan data sebesar 1%, 2,5%, dan 5% secara acak, kemudian nilai yang hilang diimputasi menggunakan GA, PSO, dan DE. Kinerja metode diukur menggunakan Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), dan Weighted Absolute Percentage Error (WAPE). Hasil menunjukkan bahwa PSO menghasilkan kinerja imputasi terbaik dengan nilai kesalahan paling rendah. Data hasil imputasi terbaik selanjutnya digunakan dalam analisis kausalitas Granger setelah dilakukan transformasi Box-Cox dengan mempertimbangkan parameter λ sebagai indikator kondisi variansi, pengujian stasioneritas terhadap rata-rata menggunakan uji Augmented Dickey-Fuller, serta penentuan lag optimal menggunakan AIC. Hasil uji menunjukkan bahwa dari 20 pasangan variabel yang dianalisis, sebanyak 19 pasangan memiliki hubungan prediktif yang signifikan, sedangkan satu pasangan, yaitu lama penyinaran matahari terhadap kelembapan, tidak signifikan. Temuan ini mengindikasikan adanya keterkaitan dinamis antarvariabel iklim di wilayah khatulistiwa serta menegaskan pentingnya metode imputasi yang akurat dalam analisis data iklim jangka panjang.

Missing data is a common issue in climate datasets that can reduce the accuracy of subsequent analyses, particularly in examining predictive relationships among variables. This study aims to evaluate the performance of metaheuristic-based imputation methods, namely Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Differential Evolution (DE), as well as to analyze predictive relationships among climate variables using the Granger causality test. The data used consist of daily climate observations from the Mempawah Climatology Station, West Kalimantan, covering the period 1990-2025 with a total of 13,149 observations. The variables analyzed include air temperature, humidity, rainfall, duration of sunshine, and wind speed. The evaluation was conducted through simulated missing data at levels of 1%, 2.5%, and 5% applied randomly, followed by imputation using GA, PSO, and DE. The performance of each method was assessed using Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Weighted Absolute Percentage Error (WAPE). The results indicate that PSO provides the best imputation performance, as reflected by the lowest error values. The best imputed dataset was subsequently used in the Granger causality analysis after applying the Box-Cox transformation, considering the lambda (λ) parameter as an indicator of variance condition, testing for stationarity in mean using the Augmented Dickey-Fuller test, and determining the optimal lag using the Akaike Information Criterion (AIC). The results show that out of 20 variable pairs analyzed, 19 exhibit significant predictive relationships, while one pair, namely sunshine duration and humidity, is not significant. These findings indicate dynamic interrelationships among climate variables in equatorial regions and highlight the importance of accurate imputation methods in long-term climate data analysis.