Peramalan Harga Saham Menggunakan Metode Hybrid Temporal Convolutional Network Dan Stockformer

Wirawan, Mochammad Afandi (2026) Peramalan Harga Saham Menggunakan Metode Hybrid Temporal Convolutional Network Dan Stockformer. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5026221127-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Pasar modal memiliki peran penting dalam mencerminkan kondisi ekonomi sekaligus menjadi sarana investasi, khususnya pada sektor perbankan seperti saham Bank Central Asia (BBCA) yang tergolong saham blue chip. Namun, pergerakan harga saham yang bersifat fluktuatif dan nonlinier menjadikan proses peramalan sebagai tantangan, terutama karena dipengaruhi oleh berbagai faktor eksternal seperti kondisi pasar yang direpresentasikan oleh Indeks Harga Saham Gabungan (IHSG). Oleh karena itu, diperlukan pendekatan yang mampu menangkap pola data deret waktu secara efektif. Penelitian ini mengusulkan model peramalan berbasis deep learning menggunakan pendekatan hybrid yang menggabungkan Temporal Convolutional Network (TCN) dan Stockformer dengan memanfaatkan data multivariat berupa harga saham BBCA dan IHSG. Eksperimen dilakukan dengan berbagai kombinasi parameter seperti TCN channels, kernel size, d_model, jumlah layer, dan learning rate serta variasi pembagian data latih dan uji. Hasil eksperimen menunjukkan bahwa seluruh model mencapai performa terbaik pada proporsi data 70:30. Pada model TCN, konfigurasi terbaik diperoleh dengan TCN channels [32, 64], kernel size 5, dan learning rate 0.005 yang menghasilkan nilai MAPE sebesar 1,069634%. Model terbaik secara keseluruhan diperoleh pada Stockformer dengan konfigurasi d_model = 64, n_head = 4, n_layers = 2, learning rate = 0.001, dan batch size = 64 yang menghasilkan nilai MAPE sebesar 1,068664% dan
=================================================================================================================================
The capital market plays an important role in reflecting economic conditions while serving as an investment medium, particularly in the banking sector, including Bank Central Asia (BBCA), which is categorized as a blue-chip stock. However, stock price movements are inherently volatile and nonlinear, making forecasting a challenging task, especially due to the influence of external factors such as overall market conditions represented by the Indonesia Composite Index (IHSG). Therefore, a forecasting approach capable of effectively capturing complex time-series patterns is required. This study proposes a deep learning-based forecasting model using a hybrid approach that combines the Temporal Convolutional Network (TCN) and Stockformer architectures with multivariate data consisting of BBCA stock prices and the IHSG. Experiments were conducted using various parameter combinations, including TCN channels, kernel size, d_model, number of layers, learning rate, and different train-test split ratios. The results indicate that all models achieved their best performance with a 70:30 train-test split. For the TCN model, the optimal configuration consisted of TCN channels of [32, 64], a kernel size of 5, and a learning rate of 0.005, resulting in a Mean Absolute Percentage Error (MAPE) of 1.069634%. The best overall performance was achieved by the Stockformer model with a configuration of d_model = 64, n_head = 4, n_layers = 2, a learning rate of 0.001, and a batch size of 64, producing a MAPE of 1.068664% and an R² value of 0.967506. In addition, the hybrid model demonstrated the most stable average performance, achieving a MAPE of 1.174757% across various parameter configurations. In the generalization test, the hybrid model outperformed Stockformer on the BBRI and TLKM datasets by producing lower MAPE values, indicating superior consistency in generalization capability. Overall, the findings demonstrate that Stockformer provides the highest forecasting accuracy and computational efficiency, whereas the hybrid model offers more consistent performance across different datasets and market conditions.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Forecasting, Multivariate, Stock, Stockformer, TCN, Multivariat, Peramalan, Saham.
Subjects:	H Social Sciences > HB Economic Theory > Economic forecasting--Mathematical models. T Technology > T Technology (General) > T174 Technological forecasting
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information System > 57201-(S1) Undergraduate Thesis
Depositing User:	Mochammad Afandi Wirawan
Date Deposited:	20 Jul 2026 06:59
Last Modified:	20 Jul 2026 06:59
URI:	http://repository.its.ac.id/id/eprint/135684

Actions (login required)

View Item