Do Technical Indicators Increase the Prediction Performance of LSTMs in Short-Term Stock Data?
16th Azerbaijan Congress of Life Engineering and Applied Sciences, Baku, Azerbaycan, 20 - 22 Eylül 2026, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Basıldığı Şehir: Baku
- Basıldığı Ülke: Azerbaycan
- Erciyes Üniversitesi Adresli: Evet
Özet
Long Short-Term Memory (LSTM) networks have been the dominant model for stock market forecasting. Several technical indicators have been extensively used in both machine learning and deep learning applications for financial prediction. This study questions the performance impact of technical indicators over LSTMs for short-term stock forecasting, where a five-day historical data window is used to predict the next day's closing value on S\&P 500 stock index data. To this aim, a framework is established, including data preparation, hyperparameter tuning with the GridSearch algorithm, multiple test experiments of two alternative approaches for performance measures, and statistical analysis. The first approach utilizes a simpler LSTM taking inputs only from closing values, while the second approach sets up a complex LSTM requiring inputs from a spectrum of technical indicator values of the last five days' data. The experimental results show that the simple LSTM achieves an average MSE of $ 3.83 \times 10^ {- 4} $, while the complex LSTM achieves an average MSE of $ 4.80 \times 10^ {- 4} $. The simpler LSTM model, relying only on closing values, gives superior predictions, where the difference is statistically significant in multiple test experiments.