Laporkan Masalah

Penerapan Metode Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise–Singular Value Decomposition–Long Short-Term Memory (ICEEMDAN–SVD–LSTM) untuk Peramalan Harga Saham

Yustian Faza, Dr. Vemmie Nastiti Lestari, S.Si., M.Sc.

2026 | Skripsi | STATISTIKA

Pergerakan harga saham yang bersifat nonlinier dan nonstasioner, serta dipengaruhi oleh berbagai faktor eksternal seperti kondisi ekonomi, sentimen pasar, dan dinamika sektoral, menuntut metode peramalan yang mampu menangkap pola kompleks dan fluktuasi yang berubah secara adaptif dalam data runtun waktu. Ketidakmampuan model konvensional dalam merepresentasikan karakteristik tersebut dapat menyebabkan penurunan akurasi prediksi dan meningkatkan risiko pengambilan keputusan investasi. Oleh karena itu, penelitian ini mengusulkan model hibrida Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise–Singular Value Decomposition–Long Short-Term Memory (ICEEMDAN–SVD–LSTM) untuk meningkatkan akurasi peramalan harga saham BBRI, UNVR, dan UNTR. ICEEMDAN digunakan untuk mendekomposisi data menjadi beberapa Intrinsic Mode Function (IMF) dan residual, kemudian dilakukan denoising menggunakan SVD berbasis matriks Hankel sebelum dimodelkan dengan LSTM untuk menangkap dependensi jangka panjang. Evaluasi menggunakan RMSE, MAE, dan MAPE menunjukkan bahwa model ICEEMDAN–SVD–LSTM secara konsisten menghasilkan nilai eror yang lebih kecil dibandingkan model pembanding, sehingga efektif dalam meningkatkan performa peramalan harga saham jangka pendek.

Stock price movements are nonlinear and non-stationary in nature and are influenced by various external factors such as economic conditions, market sentiment, and sectoral dynamics. These characteristics require forecasting methods capable of capturing complex patterns and adaptively changing fluctuations in time series data. The inability of conventional models to adequately represent these features may lead to reduced prediction accuracy and increased risk in investment decision-making. Therefore, this study proposes a hybrid model, Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise–Singular Value Decomposition–Long Short-Term Memory (ICEEMDAN–SVD–LSTM), to improve the forecasting accuracy of BBRI, UNVR, and UNTR stock prices. ICEEMDAN is employed to decompose the data into several Intrinsic Mode Functions (IMFs) and a residual component, followed by a denoising process using SVD based on a Hankel matrix before being modeled with LSTM to capture long-term dependencies. The evaluation results using RMSE, MAE, and MAPE indicate that the ICEEMDAN–SVD–LSTM model consistently produces lower error values compared to benchmark models, demonstrating its effectiveness in enhancing short-term stock price forecasting performance.

Kata Kunci : peramalan harga saham, ICEEMDAN, singular value decomposition, LSTM, runtun waktu

  1. S1-2026-497588-abstract.pdf  
  2. S1-2026-497588-bibliography.pdf  
  3. S1-2026-497588-tableofcontent.pdf  
  4. S1-2026-497588-title.pdf