Electronic Theses and Dissertation

Penelitian ini bertujuan memprediksi harga emas menggunakan metode fuzzy time series (FTS) Lee dengan membandingkan efektivitas penentuan interval melalui Formula Sturges dan metode average-based. Pengujian dilakukan untuk melihat pengaruh jumlah interval dan penerapan modifikasi threshold pada metode average-based terhadap akurasi prediksi. Hasil penelitian menunjukkan bahwa kedua pendekatan menghasilkan nilai mean absolute percentage error (MAPE) di bawah 1%, yang dikategorikan sebagai excellent forecasting. Metode average-based dengan 218 interval terbukti lebih unggul dibandingkan Formula Sturges dengan 10 interval. Hal ini divalidasi oleh nilai entropi Shannon yang lebih tinggi pada metode average-based, menunjukkan kemampuan model menangkap variasi data secara lebih detail. Selain itu, modifikasi threshold yang diterapkan pada metode average-based tahap defuzzifikasi berhasil meningkatkan akurasi dengan nilai MAPE dari 0,00962567% menjadi 0,004877558%. Hal ini mengindikasikan bahwa modifikasi threshold pada metode average-based efektif dalam mengeliminasi noise pada transisi data tunggal. Penelitian ini menyimpulkan bahwa kombinasi metode average-based dan modifikasi threshold memberikan performa optimal dalam memprediksi fluktuasi harga emas.

This study aims to predict gold prices using Lee's fuzzy time series (FTS) method by comparing the effectiveness of interval determination via Sturges' Formula and the average-based method. Testing was conducted to examine the impact of the number of intervals and the application of a threshold modification to the average-based method on prediction accuracy. The results indicate that both approaches yielded a Mean Absolute Percentage Error (MAPE) of less than 1%, categorizing them as "excellent forecasting." The average-based method with 218 intervals proved superior to Sturges' Formula with 10 intervals. This was validated by the higher Shannon entropy value associated with the average-based method, demonstrating the model's ability to capture data variations in greater detail. Furthermore, the threshold modification applied during the defuzzification stage of the average-based method successfully improved accuracy, reducing the MAPE from 0.00962567% to 0.004877558%. This indicates that the threshold modification within the average-based method is effective in eliminating noise during single-data transitions. The study concludes that the combination of the average-based method and the threshold modification delivers optimal performance in predicting gold price fluctuations.