Over-Fit or Not: That is the Question
3rd International Conference on Data, Electronics and Computing, ICDEC 2024, Kayseri, Türkiye, 18 - 20 Eylül 2024, cilt.1455 LNNS, ss.371-379, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Cilt numarası: 1455 LNNS
- Doi Numarası: 10.1007/978-981-96-7742-9_26
- Basıldığı Şehir: Kayseri
- Basıldığı Ülke: Türkiye
- Sayfa Sayıları: ss.371-379
- Anahtar Kelimeler: Long short-term memory, Machine learning, Management information system, Overfitting, Stock price prediction
- Dokuz Eylül Üniversitesi Adresli: Evet
Özet
Stock performance models are tightly linked to recent data. The ability to accurately predict stock prices has many advantages. It helps investors better understand future market trends and price movements and make the right decisions. By analyzing short- and long-term market movements, it can guide fast and strategic decisions. Machine learning algorithms are among the methods successfully used in stock price forecasting. Within the scope of the study, Linear Regression, Multilayer Perceptron (MLP), Long Short-Term Memory (LSTM) and Support Vector Regression (SVR) models were used to estimate future price movements by analysing historical data and the best performance was obtained in the LSTM model (R-Square = 0.997). However, the high R-Squared value indicates that the model overfits the data set, the model loses its generalisation ability and thus there is a danger of overfitting. Therefore, necessary precautions were taken against possible overfitting situations and the model was ensured to make reliable predictions with various methods.