View Item 
      •   IPB Repository
      • Dissertations and Theses
      • Undergraduate Theses
      • UT - Faculty of Mathematics and Natural Sciences
      • UT - Physics
      • View Item
      •   IPB Repository
      • Dissertations and Theses
      • Undergraduate Theses
      • UT - Faculty of Mathematics and Natural Sciences
      • UT - Physics
      • View Item
      JavaScript is disabled for your browser. Some features of this site may not work without it.

      Pengembangan Model LSTM untuk Prediksi Harga Saham Migas Menggunakan Optimasi Momentum Hibrida Berbasis Newtonian dan L-BFGS

      Thumbnail
      View/Open
      Cover (2.966Mb)
      Fulltext (9.831Mb)
      Lampiran (3.441Mb)
      Date
      2025
      Author
      Sukmayadi, Cecep Hardi
      Hardhienata, Hendradi
      Alatas, Husin
      Metadata
      Show full item record
      Abstract
      Penelitian ini mengevaluasi model Long Short-Term Memory (LSTM) untuk prediksi harga saham migas dengan optimasi berbasis momentum: gradient descent, gradient descent analogi Newtonian, dan hibrida. Optimasi Newtonian memodelkan pembaruan parameter sebagai gerak partikel dalam fluida kental dengan gaya hambat. Optimasi hibrida terdiri atas dua fase, yaitu fase awal menggunakan akselerasi Nesterov untuk mempercepat konvergensi, dilanjutkan L-BFGS untuk penyempurnaan. Dua optimasi pertama digunakan sebagai baseline. Model diuji pada data historis tiga emiten migas: PT Medco Energi Internasional Tbk., PT Perusahaan Gas Negara Tbk., dan PT Elnusa Tbk. Secara umum, semua model mampu menangkap pola tren historis dengan akurasi yang baik, dan pendekatan hibrida menunjukkan performa yang paling kompetitif pada salah satu emiten (MAPE 1,84%). Meskipun hasilnya belum konsisten di seluruh pengujian, pendekatan ini menunjukkan potensi sebagai alternatif teknik optimasi. Namun, tantangan yang berkaitan dengan kompatibilitas terhadap graph mode TensorFlow masih dapat membatasi efisiensi dan skalabilitas saat implementasi.
       
      This study evaluates Long Short-Term Memory (LSTM) models for oil and gas stock price prediction using momentum-based optimizations: gradient descent, Newtonian analogy gradient descent, and hybrid optimization. Newtonian optimization models parameter updates as particle motion in a viscous fluid with drag. Hybrid optimization consists of two phases: an initial phase using Nesterov acceleration to accelerate convergence, followed by L-BFGS for refinement. The first two optimizations are used as baselines. The models are tested on historical data from three oil and gas issuers: PT Medco Energi Internasional Tbk., PT Perusahaan Gas Negara Tbk., and PT Elnusa Tbk. In general, all models can capture historical trend patterns with good accuracy, and the hybrid approach shows the most competitive performance on one issuer (MAPE 1.84%). Although the results are not consistent across tests, this approach shows potential as an alternative optimization technique. However, technical challenges remain, particularly related to compatibility with TensorFlow graph mode, which can limit efficiency and scalability during implementation.
       
      URI
      http://repository.ipb.ac.id/handle/123456789/171767
      Collections
      • UT - Physics [1230]

      Copyright © 2020 Library of IPB University
      All rights reserved
      Contact Us | Send Feedback
      Indonesia DSpace Group 
      IPB University Scientific Repository
      UIN Syarif Hidayatullah Institutional Repository
      Universitas Jember Digital Repository
        

       

      Browse

      All of IPB RepositoryCollectionsBy Issue DateAuthorsTitlesSubjectsThis CollectionBy Issue DateAuthorsTitlesSubjects

      My Account

      Login

      Application

      google store

      Copyright © 2020 Library of IPB University
      All rights reserved
      Contact Us | Send Feedback
      Indonesia DSpace Group 
      IPB University Scientific Repository
      UIN Syarif Hidayatullah Institutional Repository
      Universitas Jember Digital Repository