SANJAYA, Daniel Robi and Surarso, Bayu and Tarno, Tarno (2025) MODIFIKASI MODEL LONG SHORT-TERM MEMORY DALAM MENINGKATKAN AKURASI PREDIKSI PADA DATA TIME SERIES. Masters thesis, UNIVERSITAS DIPONEGORO.

	Text COVER TESIS.pdf Restricted to Repository staff only Download (1MB)
	Text BAB I.pdf Download (319kB)
	Text BAB II.pdf Download (1MB)
	Text BAB III.pdf Restricted to Repository staff only Download (612kB)
	Text BAB IV.pdf Restricted to Repository staff only Download (4MB)
	Text BAB V.pdf Restricted to Repository staff only Download (255kB)
	Text DAFTAR PUSTAKA.pdf Download (306kB)
	Text LAMPIRAN.pdf Restricted to Repository staff only Download (1MB)

Abstract

Penelitian dilatarbelakangi pentingnya peramalan yang akurat pada data time series dalam sebuah harga saham, dalam mendukung keputusan investasi yang lebih baik. Mengingat tingginya fluktuasi harga saham yang dipengaruhi oleh berbagai faktor dan model prediksi yang kurang akurat, diperlukan pengembangan model prediksi yang lebih akurat dan efisien dalam membantu investor mengambil keputusan investasi. Penelitian ini bertujuan untuk meningkatkan akurasi model deep learning LSTM dalam prediksi harga saham. Salah satu upaya untuk menanggulangi permasalahan tersebut yaitu modifikasi dan pengoptimalan model LSTM dengan penyederhanaan hyperparameter dan penambahan layer dense, serta menerapkan optimasi Adam untuk menghasilkan prediksi yang lebih akurat dibandingkan dengan metode prediksi lainnya. Algoritma yang digunakan dalam penelitian ini adalah salah satu model deep learning Long Short-Term Memory (LSTM), karena memiliki kelebihan dalam memprediksi data time series, serta mampu mempelajari pola dan ketergantungan data dalam jangka panjang, sehingga dapat menghasilkan prediksi yang lebih akurat dibandingkan metode sederhana. Hasil penelitian ini menunjukan bahwa model LSTM yang telah dimodifikasi mengalami peningkatan dan memiliki akurasi lebih baik dibandingkan metode LSTM sebelum dimodifikasi dan metode prediksi sederhana seperti Weighted Moving Average (WMA), Simple Exponential Smoothing (SES), dan Facebook Prophet. Hasil evaluasi akurasi MAPE pada model LSTM sebelum dimodifikasi sebesar 7,64% pada data training, dan 2,65% testing, dan 3,51% (training), 1,65% (testing) pada model LSTM yang dimodifikasi, serta 4,30% (training), 2,10% (testing) model SES, dan 8,73% (training), 3,63% (testing) model WMA, dan 10,04% (training), 5.53% (testing) pada model Facebook Prophet.
Kata Kunci: Modifikasi, LSTM, Metode Sederhana, Akurasi, Saham, Time Series

The research is motivated by the importance of accurate forecasting on time series data in a stock price, in supporting better investment decisions. Given the high fluctuations in stock prices that are influenced by various factors and inaccurate prediction models, it is necessary to develop a more accurate and efficient prediction model in helping investors make investment decisions. This research aims to improve the accuracy of the deep learning LSTM model in predicting stock prices. One of the efforts to overcome these problems is to modify and optimize the LSTM model by simplifying the hyperparameters and adding dense layers, and applying Adam's optimization to produce more accurate predictions compared to other prediction methods. The algorithm used in this research is one of the Long Short-Term Memory (LSTM) deep learning models, because it has advantages in predicting time series data, and is able to learn patterns and data dependencies in the long term, so that it can produce predictions that are more accurate than simple methods. The results of this study show that the modified LSTM model has improved and has better accuracy than the LSTM method before modification and simple prediction methods such as Weighted Moving Average (WMA), Simple Exponential Smoothing (SES), and Facebook Prophet. The MAPE accuracy evaluation results on the LSTM model before being modified were 7,64% on training data, and 2,65% testing, and 3,51% (training), 1,65% (testing) on the modified LSTM model, as well as 4.30% (training), 2,10% (testing) SES model, and 8,73% (training), 3,63% (testing) WMA model, and 10,04% (training), 5,53% (testing) on the Facebook Prophet model.
Keywords: Modification, LSTM, Simple Method, Accuracy, Stocks, Time Series

Actions (login required)

View Item