Peramalan Harga Saham PT Indofood Sukses Makmur Tbk. (INDF) Menggunakan Metode Temporal Convolutional Network (TCN) Dan Bidirectional Gated Recurrent Unit (BiGRU)

Amartika, Nida Aulia (2026) Peramalan Harga Saham PT Indofood Sukses Makmur Tbk. (INDF) Menggunakan Metode Temporal Convolutional Network (TCN) Dan Bidirectional Gated Recurrent Unit (BiGRU). Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5026221095-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Prediksi data runtun waktu yang akurat menjadi krusial dalam pengambilan keputusan strategis. Penelitian ini mengevaluasi dan mengoptimalkan performa arsitektur hybrid TCN-BiGRU dibandingkan dengan model tunggal TCN dan BiGRU melalui variasi proporsi data (60:20:20, 70:15:15, dan 80:10:10) serta hyperparameter tuning yang sistematis. Hasil penelitian menunjukkan bahwa arsitektur hybrid TCN-BiGRU mencapai performa puncak pada skenario 60:20:20 dengan konfigurasi Filter 64, Kernel 3, GRU Units 64, Batch Size 8, dan Learning Rate 0,0015, menghasilkan MAPE sebesar 1,0565% pada epoch ke-35. Analisis komparatif mengungkapkan adanya trade-off signifikan antara kompleksitas model dan efisiensi waktu pemrosesan. Sebagai kontribusi praktis, sistem visualisasi berbasis website dengan arsitektur back-end Flask dikembangkan untuk memudahkan interpretasi hasil prediksi secara komparatif. Kesimpulan penelitian ini menegaskan bahwa pemilihan arsitektur model dan proporsi data harus disesuaikan dengan kebutuhan implementasi, antara prioritas ketepatan prediksi atau efisiensi sumber daya komputasi.
==================================================================================================================================
Accurate time series forecasting is crucial for strategic decision-making. This research evaluates and optimizes the TCN-BiGRU hybrid architecture compared to individual TCN and BiGRU models by analyzing various data proportions (60:20:20, 70:15:15, and 80:10:10) and systematic hyperparameter tuning. Results demonstrate that the TCN-BiGRU hybrid architecture achieves peak performance under the 60:20:20 data split with a configuration of 64 Filters, a Kernel size of 3, 64 GRU Units, a Batch Size of 8, and a Learning Rate of 0.0015, yielding a MAPE of 1.0565% at the 35th epoch. A comparative analysis reveals a significant trade-off between architectural complexity and processing efficiency. As a practical contribution, a Flask-based web visualization system was developed to facilitate the comparative interpretation of forecast results. The study concludes that the selection of model architecture and data proportion must be aligned with specific implementation needs, balancing the priority between prediction precision and computational resource efficiency.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	TCN-BiGRU, Prediksi Time Series, Hyperparameter Tuning, Trade-off Komputasi, Visualisasi Website. TCN-BiGRU, Time Series Forecasting, Hyperparameter Tuning , Computational Trade-off, Web Visualization.
Subjects:	Q Science > QA Mathematics > QA276 Mathematical statistics. Time-series analysis. Failure time data analysis. Survival analysis (Biometry) Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information System > 57201-(S1) Undergraduate Thesis
Depositing User:	Nida Aulia Amartika
Date Deposited:	29 Jul 2026 08:24
Last Modified:	29 Jul 2026 08:24
URI:	http://repository.its.ac.id/id/eprint/139726

Actions (login required)

View Item