@phdthesis{repoupi137822,
          school = {Universitas Pendidikan Indonesia},
           title = {PERBANDINGAN KINERJA MODEL SGDREGRESSOR BERBASIS INCREMENTAL LEARNING UNTUK PEMODELAN HARGA SAHAM: Studi Kasus TLKM.JK},
            note = {https://scholar.google.com/citations?user=elEleAkAAAAJ\&hl=id

ID SINTA Dosen Pembimbing:
Herbert Siregar: 5991008
Ani Anisyah: 6786982},
           month = {August},
            year = {2025},
        abstract = {Pemodelan harga saham memiliki peran penting dalam mendukung pengambilan keputusan keuangan karena membantu investor dan pemangku kepentingan menilai tren pasar, peluang investasi, dan potensi risiko. Mengingat pasar saham yang bersifat dinamis dan mudah berubah, diperlukan model yang tidak hanya akurat, tetapi juga adaptif terhadap data terbaru. Model batch learning tradisional sering digunakan karena mampu menghasilkan akurasi tinggi ketika dilatih pada dataset lengkap, namun memiliki keterbatasan dalam pembaruan saat data baru tersedia. Penelitian ini bertujuan membandingkan kinerja model SGDRegressor berbasis incremental learning dengan pendekatan batch learning dalam pemodelan harga saham PT Telkom Indonesia Tbk (TLKM) menggunakan data historis periode 2015-2024 dari Yahoo Finance. Tiga skenario pembelajaran diuji, yaitu incremental learning tahunan, dua tahunan, serta batch learning. Evaluasi dilakukan menggunakan lima metrik performa, yaitu MSE, RMSE, MAE, MAPE, dan R2. Hasil penelitian menunjukkan bahwa batch learning memberikan akurasi tertinggi (RMSE 18,55; MAPE 0,49; R2 0,9969). Skenario incremental learning dua tahunan menunjukkan keseimbangan terbaik antara akurasi dan adaptabilitas (rata-rata R2 0,9658), sedangkan pembaruan tahunan lebih adaptif namun sedikit kurang akurat akibat ukuran data latih yang lebih kecil. Temuan ini menegaskan bahwa SGDRegressor berbasis incremental learning merupakan alternatif efisien dan praktis karena mampu mempertahankan kinerja kompetitif sekaligus mendukung pembaruan model secara berkelanjutan seiring tersedianya data baru.

Stock price modeling plays a crucial role in supporting financial decision-making as it helps investors and stakeholders assess market trends, investment opportunities, and potential risks. Given the dynamic and volatile nature of the stock market, it is essential to employ models that are not only accurate but also adaptive to new data. Traditional batch learning models are widely used due to their ability to achieve high accuracy when trained on complete datasets; however, they have limitations in updating when new data become available. This study aims to compare the performance of the SGDRegressor model based on incremental learning with the batch learning approach in modeling the stock price of PT Telkom Indonesia Tbk (TLKM) using historical data from 2015-2024 obtained from Yahoo Finance. Three learning scenarios were tested: annual incremental learning, biennial incremental learning, and batch learning. The evaluation was conducted using five performance metrics: MSE, RMSE, MAE, MAPE, and R2. The results show that batch learning achieved the highest accuracy (RMSE 18.55; MAPE 0.49; R2 0.9969). The biennial incremental learning scenario demonstrated the best balance between accuracy and adaptability (average R2 0.9658), while annual updates were more adaptive but slightly less accurate due to smaller training data size. These findings highlight that the SGDRegressor with incremental learning is an efficient and practical alternative, capable of maintaining competitive performance while supporting continuous model updates as new data become available.},
             url = {https://repository.upi.edu/},
          author = {Rizal Teddyansyah, - and Herbert Siregar, - and Ani Anisyah, -},
        keywords = {Batch Learning, Incremental Learning, Machine Learning, Pemodelan Harga Saham, SGDRegressor

Batch Learning, Incremental Learning, Machine Learning SGDRegressor, Stock Price Modeling}
}