Penerapan Metode Hybrid Complete Ensemble Empirical Mode Decomposition with Adaptive Noise-Recurrent Neural Networks Pada Peramalan Harga Saham

Eraswati, Kadek Imelda Anindra (2025) Penerapan Metode Hybrid Complete Ensemble Empirical Mode Decomposition with Adaptive Noise-Recurrent Neural Networks Pada Peramalan Harga Saham. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5006211010-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (10MB) | Request a copy

Abstract

Investasi saham merupakan salah satu bentuk investasi yang bertujuan memperoleh keuntungan di masa depan serta melindungi nilai aset dari dampak inflasi. Namun, harga saham bersifat fluktuatif dan dipengaruhi oleh berbagai faktor sehingga diperlukan metode peramalan yang akurat guna meminimalkan risiko investasi. Penelitian ini menerapkan metode hybrid untuk peramalan harga saham dengan menggabungkan metode dekomposisi Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) dan metode peramalan Recurrent Neural Networks (RNN), yaitu Long Short Term Memory (LSTM) dan Gated Recurrent Unit (GRU). Data yang digunakan adalah harga saham PT Alamtri Resources Indonesia Tbk (ADRO) selama periode 1 Januari 2022 hingga 30 Desember 2024. Hasil dekomposisi CEEMDAN menghasilkan 7 komponen Intrinsic Mode Functions (IMF) dan 1 komponen residual yang dilakukan peramalan menggunakan dua pendekatan rekonstruksi. Pendekatan pertama dilakukan dengan menjumlahkan seluruh hasil peramalan IMF dan residual berdasarkan kombinasi input lag dan hyperparameter yang sama. Pendekatan kedua menjumlahkan hasil peramalan IMF dan residual berdasarkan kombinasi hyperparameter terbaik yang dipilih secara terpisah untuk setiap komponen, namun tetap menggunakan kombinasi input yang sama. Hasil evaluasi menunjukkan bahwa pendekatan kedua pada CEEMDAN-LSTM menghasilkan performa terbaik dalam prediksi data historis dengan MAE sebesar 66,7775, RMSE sebesar 128,9119, dan MAPE sebesar 2,3101%. Sementara itu, CEEMDAN-GRU menghasilkan performa terbaik pada pendekatan pertama dengan MAE sebesar 67,0277, RMSE sebesar 131,0742, dan MAPE sebesar 2,3143%. Berdasarkan hasil tersebut, metode hybrid CEEMDAN-LSTM terbukti lebih unggul dalam melakukan prediksi harga saham historis ADRO. Meskipun demikian, kedua metode sama-sama memenuhi kriteria akurasi sangat baik dengan nilai MAPE<10%. Metode CEEMDAN-LSTM kemudian digunakan untuk melakukan peramalan harga saham selama dua puluh hari ke depan, yang menunjukkan fluktuasi harga saham dalam kisaran Rp3.690 hingga Rp3.880 per lembar. Pergerakan harga menunjukkan tren menurun di awal periode, diikuti peningkatan secara bertahap, dan mencapai puncak pada pertengahan hingga akhir periode, yang mencerminkan stabilitas pergerakan harga saham secara umum.
================================================================================================================================
Stock investment is one type of investment aimed at generating future profits and protecting asset value from the impact of inflation. However, stock prices are inherently volatile and influenced by various factors, thus requiring accurate forecasting methods to minimize investment risks. This study applies a hybrid forecasting method by combining Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and Recurrent Neural Networks (RNN), namely Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU). The dataset used consists of daily stock prices of PT Alamtri Resources Indonesia Tbk (ADRO) from January 1, 2022, to December 30, 2024. The CEEMDAN decomposition produced seven Intrinsic Mode Functions (IMFs) and one residual component, each of which was forecasted using two reconstruction approaches. The first approach reconstructs the signal by summing the forecasts of all IMFs and the residual using the same input lag and hyperparameter combination. The second approach reconstructs the signal by summing the forecasts of each component using the best hyperparameter combination selected separately for each IMF and residual, while maintaining the same input configuration. Evaluation results show that the second approach using CEEMDAN-LSTM achieved the best performance in historical data prediction, with MAE of 66.7775, RMSE of 128.9119, and MAPE of 2.3101%. Meanwhile, CEEMDAN-GRU performed best under the first approach with MAE of 67.0277, RMSE of 131.0742, and MAPE of 2.3143%. These results indicate that the CEEMDAN-LSTM hybrid method outperforms CEEMDAN-GRU in predicting historical stock prices of ADRO. Nevertheless, both methods meet the criteria for excellent forecasting accuracy with MAPE < 10%. The CEEMDAN-LSTM method was then used to forecast stock prices for the next twenty days, revealing price fluctuations within the range of IDR 3,690 to IDR 3,880 per share. The price movement shows a downward trend at the beginning of the period, followed by a gradual increase and reaching a peak in the mid to late period, reflecting overall stability in stock price movement.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	CEEMDAN, GRU, LSTM, Metode Hybrid, Peramalan Harga Saham, CEEMDAN, GRU, Hybrid Method, LSTM, Stock Price Forecasting.
Subjects:	H Social Sciences > HA Statistics > HA30.3 Time-series analysis H Social Sciences > HG Finance > HG4910 Investments H Social Sciences > HG Finance > HG4915 Stocks--Prices Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Actuaria > 94203-(S1) Undergraduate Thesis
Depositing User:	Kadek Imelda Anindra Eraswati
Date Deposited:	14 Jul 2025 03:15
Last Modified:	14 Jul 2025 03:15
URI:	http://repository.its.ac.id/id/eprint/119664

Actions (login required)

View Item