PERBANDINGAN MODEL GAUSSIAN PROCESS REGRESSION DAN LONG SHORT-TERM MEMORY DALAM PREDIKSI HARGA SAHAM LIMA BANK BUMN PERIODE 2018-2024

Kristananda, Raja Valentino (2025) PERBANDINGAN MODEL GAUSSIAN PROCESS REGRESSION DAN LONG SHORT-TERM MEMORY DALAM PREDIKSI HARGA SAHAM LIMA BANK BUMN PERIODE 2018-2024. Undergraduate thesis, Universitas Pembangunan Nasional Veteran Jawa Timur.

	Text (Cover) 21083010068_Cover.pdf Download (677kB)
	Text (Bab 1) 21083010068_Bab 1.pdf Download (30kB)
	Text (Bab 2) 21083010068_Bab 2.pdf Restricted to Registered users only until 19 June 2028. Download (353kB)
	Text (Bab 3) 21083010068_Bab 3.pdf Restricted to Registered users only until 19 June 2028. Download (115kB)
	Text (Bab 4) 21083010068_Bab 4.pdf Restricted to Registered users only until 19 June 2028. Download (3MB)
	Text (Bab 5) 21083010068_Bab 5.pdf Download (82kB)
	Text (Daftar Pustaka) 21083010068_Daftar Pustaka.pdf Download (161kB)
	Text (Lampiran) 21083010068_Lampiran.pdf Restricted to Registered users only until 19 June 2028. Download (128kB)

Abstract

The stock price prediction of five BUMN banks is crucial for the stability of Indonesia's economy due to their strategic role as stabilizers of government policies. The main issue lies in the limitations of traditional methods in capturing nonlinear patterns and the absence of a comparative study between probabilistic non-parametric approaches and deep learning for BUMN stocks. The urgency of this problem arises from stock market volatility and the need for accurate predictions to support investment decisions and economic policy-making. This study compares Gaussian Process Regression (GPR) with four kernels and Long Short-Term Memory (LSTM) across four architectural variations using data from 2018-2024 (1,475 data points) with an 80%-10%-10% split, and evaluates using MAE, RMSE, MAPE, and out-of-sample testing. The significance of GPR lies in its ability to model uncertainty and nonlinear patterns, while LSTM excels in handling time-series data with long-term temporal patterns. The innovation of this research is the comprehensive comparative study between GPR with various kernels and LSTM in the context of Indonesian BUMN stocks with out-of-sample evaluation. The results show that the GPR Matern kernel outperforms for four stocks with the lowest RMSE: BBRI (73.94), BMRI (88.34), BBNI (70.38), BBTN (16.29), and MAPE ranging from 0.98% to 1.09%. The GPR Linear kernel is optimal for BRIS with an RMSE of 35.49 and MAPE of 1.47%. LSTM performed lower, with the highest RMSE of 200.07 for BMRI and MAPE ranging from 1.47% to 2.57%. The five-year historical data period produced the lowest MAPE of 1.27% for BMRI stock. Out-of-sample evaluation shows that GPR maintains accuracy with a MAPE of 3.15% for 30-day predictions, while LSTM experiences significant degradation with MAPE reaching up to 12.96%. The findings confirm the superiority of GPR in modeling the uncertainty of the BUMN stock market, providing empirical evidence for the development of an investment support system based on probabilistic modeling.

Item Type:

Thesis (Undergraduate)

Contributors:

Contribution	Contributors	NIDN/NIDK	Email
Thesis advisor	Prasetya, Dwi Arman	NIDN0005128001	arman.prasetya.sada@upnjatim.ac.id
Thesis advisor	Hindrayani, Kartika Maulida	NIDM0009099205	kartika.maulida.ds@upnjatim.ac.id

Subjects:

H Social Sciences > HA Statistics
Q Science > QA Mathematics
Q Science > QA Mathematics > QA76.6 Computer Programming

Divisions:

Faculty of Computer Science > Departemen of Data Science

Depositing User:

Raja Valentino Kristananda

Date Deposited:

20 Jun 2025 02:25

Last Modified:

20 Jun 2025 02:25

URI:

https://repository.upnjatim.ac.id/id/eprint/38570