Peramalan Harga Saham PT Indofood Sukses Makmur Tbk. (INDF) Menggunakan Metode Hybrid Convolutional Neural Network (CNN) dan Bidirectional Long Short-Term Memory (BiLSTM)

Nuha, Talitha Firyal Ghina (2026) Peramalan Harga Saham PT Indofood Sukses Makmur Tbk. (INDF) Menggunakan Metode Hybrid Convolutional Neural Network (CNN) dan Bidirectional Long Short-Term Memory (BiLSTM). Other thesis, Institut Teknologi Sepuluh Nopember.

Text 031_Proposal TA SemHas_Talitha Firyal Ghina Nuha.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Harga saham PT Indofood Sukses Makmur Tbk. (INDF) bergerak dengan pola yang kompleks akibat pengaruh volatilitas pasar, perubahan sentimen investor, dan dinamika ekonomi yang terus bergeser, sehingga menyulitkan metode konvensional dalam menangkap karakter non-linear pada data finansial. Tugas Akhir ini membangun model hybrid Convolutional Neural Network dan Bidirectional Long Short-Term Memory (CNN-BiLSTM) untuk meramalkan harga saham INDF, serta mengevaluasi akurasi dan kinerjanya dibandingkan dengan model CNN dan BiLSTM tunggal. Data historis harga penutupan harian INDF selama sepuluh tahun yang diperoleh dari investing.com diproses melalui tahap interpolasi missing value, deteksi outlier, dan normalisasi Min-Max Scaling, kemudian dibagi ke dalam tiga skenario (80:10:10, 70:15:15, dan 60:20:20) dengan pendekatan sliding window 30 hari. Arsitektur hybrid CNN-BiLSTM dirancang dengan mengombinasikan kemampuan CNN dalam mengekstraksi pola lokal dan kemampuan BiLSTM dalam mempelajari hubungan temporal dua arah, dilatih menggunakan fungsi loss Mean Squared Error dengan optimizer Adam, serta dioptimasi melalui hyperparameter tuning menggunakan grid search. Hasil pengujian menunjukkan bahwa model hybrid CNN-BiLSTM terbaik, pada skenario pembagian data 60:20:20, menghasilkan kinerja tertinggi di antara ketiga model dengan nilai MAPE 1,1737%, MAE 84,77, RMSE 126,18, dan R² 0,9707, sedikit lebih unggul dibandingkan CNN deangn MAPE 1,1824% dan BiLSTM dengan MAPE 1,2743%, dengan waktu pelatihan yang juga paling efisien di antara ketiga model. Hasil peramalan 30 hari ke depan turut menunjukkan kecenderungan tren menurun pada ketiga model dengan besaran yang bervariasi antar model. Model hybrid CNN-BiLSTM juga divisualisasikan melalui sistem berbasis web untuk memudahkan interpretasi hasil prediksi oleh pengguna. Temuan ini menunjukkan bahwa pendekatan hybrid CNN-BiLSTM mampu menghasilkan prediksi harga saham INDF dengan akurasi tinggi, sekaligus memberikan kontribusi bagi pengembangan teknik peramalan harga saham berbasis deep learning serta manfaat praktis bagi investor dalam pengambilan keputusan investasi.
=================================================================================================================================
The stock price of PT Indofood Sukses Makmur Tbk. (INDF) follows a complex pattern due to market volatility, shifts in investor sentiment, and constantly changing economic dynamics, making it difficult for conventional methods to capture the nonlinear nature of financial data. This thesis develops a hybrid Convolutional Neural Network and Bidirectional Long Short-Term Memory (CNN-BiLSTM) model to forecast INDF stock prices, and evaluates its accuracy and performance compared to standalone CNN and BiLSTM models. The ten years of historical daily closing price data for INDF, obtained from investing.com, was processed through stages of missing value interpolation, outlier detection, and Min-Max Scaling normalization, then divided into three scenarios (80:10:10, 70:15:15, and 60:20:20) using a 30-day sliding window approach. The hybrid CNN-BiLSTM architecture was designed by combining the CNN’s ability to extract local patterns with the BiLSTM’s ability to learn bidirectional temporal relationships. It was trained using the Mean Squared Error loss function with the Adam optimizer and optimized through hyperparameter tuning using grid search. The test results show that the best hybrid CNN-BiLSTM model, under a 60:20:20 data split, achieved the highest performance among the three models with a MAPE of 1.1737%, an MAE of 84.77, an RMSE of 126.18, and an R² of 0.9707, slightly outperforming the CNN with an MAPE 1.1824% and BiLSTM with an MAPE 1.2743%, while also having the most efficient training time among the three models. The 30-day forecast results also show a downward trend across all three models, with the magnitude of the decline varying among them. The hybrid CNN-BiLSTM model was also visualized via a web-based system to facilitate users’ interpretation of the prediction results. These findings indicate that the hybrid CNN-BiLSTM approach is capable of generating highly accurate predictions of INDF stock prices, while also contributing to the development of deep learning-based stock price forecasting techniques and providing practical benefits for investors in their investment decision-making.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	CNN-BiLSTM, Deep Learning, Harga Saham, Peramalan, Time Series. CNN-BiLSTM, Deep Learning, Forecasting, Stock Prices, Time Series.
Subjects:	T Technology > T Technology (General) > T174 Technological forecasting
Divisions:	Faculty of Information Technology > Information System > 57201-(S1) Undergraduate Thesis
Depositing User:	Talitha Firyal Ghina Nuha
Date Deposited:	24 Jul 2026 21:49
Last Modified:	24 Jul 2026 21:49
URI:	http://repository.its.ac.id/id/eprint/137333

Actions (login required)

View Item