Perbandingan Kinerja Model DSD-LSTM dan DSD-GRU dalam Peramalan Harga Saham Sektor Keuangan Berdasarkan Clustering Time Series dengan K-Medoids-DTW

Marfilyamin, Rohviska Novilita (2026) Perbandingan Kinerja Model DSD-LSTM dan DSD-GRU dalam Peramalan Harga Saham Sektor Keuangan Berdasarkan Clustering Time Series dengan K-Medoids-DTW. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5006221015-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (7MB) | Request a copy

Abstract

Pasar saham mengalami pertumbuhan yang signifikan sehingga diperlukan metode yang mampu mengolah data menjadi lebih sederhana dan menghasilkan prediksi harga yang akurat. Penelitian ini bertujuan untuk membandingkan kinerja pendekatan Dense-Sparse-Dense (DSD) pada model Long Short-Term Memory (LSTM) dan Gated Unit Recurrent (GRU) dalam prediksi harga saham sektor keuangan berdasarkan hasil clustering time series. Proses analisis diawali dengan mengelompokkan 41 data harga penutupan harian saham perusahaan sektor keuangan yang tercatat di Papan Utama Bursa Efek Indonesia periode 3 Januari 2022 - 31 Juli 2025 menggunakan metode K-Medoids dengan ukuran jarak Dynamic Time Warping (DTW). Evaluasi Silhouette Score (0.618), Davies-Bouldin Index (DBI) (0,478), dan Pseudo-F (75,304) menunjukkan bahwa enam klaster merupakan jumlah klaster optimal dengan medoid masing-masing. Medoid tiap klaster kemudian dimodelkan dengan DSD-LSTM dan DSD-GRU. Hasil pemodelan menunjukkan bahwa DSD-GRU unggul pada 5 dari 6 klaster berdasarkan metrik Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), dan Root Mean Squared Error (RMSE), sedangkan DSD-LSTM unggul pada satu klaster. Model terbaik diaplikasikan ke seluruh anggota klaster melalui proses fine tuning menghasilkan nilai rata-rata MAPE di bawah 3% pada sebagian besar klaster yang menunjukkan tingkat akurasi prediksi sangat baik dan model mampu menangkap pola pergerakan harga pada masing-masing klaster secara efektif. Kemudian dilanjutkan peramalan untuk periode 1 Agustus 2025 – 29 Agustus 2025 dengan hasil klaster 1, 2, 4, dan 5 cenderung meningkat, klaster 3 melemah, dan klaster 6 fluktuatif tajam sehingga hasil tersebut dapat mendukung pengambilan keputusan investasi yang lebih tepat.
======================================================================================================================================
The stock market has experienced significant growth, increasing the need for methods capable of simplifying complex data and producing accurate price predictions. This study aims to compare the performance of the Dense-Sparse-Dense (DSD) approach applied to Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models in forecasting stock prices in the financial sector based on time-series clustering results. The analysis begins by grouping 41 daily closing price series of financial-sector companies listed on the Main Board of the Indonesia Stock Exchange for the period of January 3, 2022–July 31, 2025 using the K-Medoids method with Dynamic Time Warping (DTW) distance. The evaluation metrics Silhouette Score (0.618), Davies-Bouldin Index (0.478), and Pseudo-F (75.304) indicate that six clusters represent the optimal number of clusters, each with its respective medoid. The medoids of each cluster were then modeled using DSD-LSTM and DSD-GRU. The modeling results show that DSD-GRU outperforms in 5 out of 6 clusters based on Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Root Mean Squared Error (RMSE), while DSD-LSTM performs better in one cluster. The best-performing model was subsequently applied to all cluster members through fine-tuning, yielding average MAPE values below 3% in most clusters, indicating excellent prediction accuracy and the model’s effectiveness in capturing price movement patterns within each cluster. Finally, forecasting for the period of August 1, 2025–August 29, 2025 reveals that clusters 1, 2, 4, and 5 tend to increase, cluster 3 weakens, and cluster 6 exhibits sharp fluctuations. These findings provide valuable insights to support more informed investment decision-making.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Clustering, Dense-Sparse-Dense, Dynamic Time Warping, GRU, K-Medoids LSTM
Subjects:	Q Science > QA Mathematics > QA276 Mathematical statistics. Time-series analysis. Failure time data analysis. Survival analysis (Biometry) Q Science > QA Mathematics > QA278.55 Cluster analysis Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) Q Science > QA Mathematics > QA76.9.D343 Data mining. Querying (Computer science) Q Science > QA Mathematics > QA278 Cluster Analysis. Multivariate analysis. Correspondence analysis (Statistics)
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Actuaria > 94203-(S1) Undergraduate Thesis
Depositing User:	Rohviska Novilita Marfilyamin
Date Deposited:	12 Jan 2026 05:26
Last Modified:	12 Jan 2026 05:26
URI:	http://repository.its.ac.id/id/eprint/129482

Actions (login required)

View Item