Related Experiment Video
Updated: Apr 23, 2026
07:34
A Simple Stimulatory Device for Evoking Point-like Tactile Stimuli: A Searchlight for LFP to Spike Transitions
Published on: March 25, 2014
9.3K
Forecasting Nasdaq stock exchange time series using an improved recurrent spiking Pi-Sigma artificial neural network
Erol Egrioglu1, Eren Bas2, Gulsen Albayrak2
1Department of Data Science and Analytics, Faculty of Arts and Science, Giresun University, 28200, Giresun, Turkey. erol.egrioglu@giresun.edu.tr.
Scientific Reports
|April 21, 2026
Summary
This study introduces a novel artificial neural network for improved forecasting. The new model, incorporating multiplicative and additive neurons with dynamic particle swarm optimization, demonstrates successful performance in Nasdaq stock time series predictions.
Area of Science:
- Artificial Intelligence
- Machine Learning
- Time Series Analysis
Background:
- Artificial neural networks (ANNs) excel in nonlinear modeling and forecasting.
- Neuron model variations significantly impact ANN forecasting performance.
- Existing methods require enhancement for complex time series data.
Purpose of the Study:
- To introduce a novel ANN architecture for enhanced forecasting.
- To integrate multiplicative and additive neuron models with exponential smoothing feedback.
- To develop a dynamic particle swarm optimization training algorithm.
Main Methods:
- Developed a new ANN architecture combining multiplicative and additive neuron models.
- Implemented feedback logic from exponential smoothing methods.
- Utilized particle swarm optimization with a dynamic fitness function prioritizing recent data.
- Conducted statistical hypothesis tests for performance evaluation.
Main Results:
- The proposed ANN demonstrated successful forecasting performance on Nasdaq stock exchange time series.
- Empirical results confirmed the effectiveness of the novel architecture.
- The dynamic fitness function in PSO training improved recent observation weighting.
Conclusions:
- The novel ANN architecture offers a promising alternative for time series forecasting.
- Integration of specific neuron models and training algorithms enhances predictive accuracy.
- The proposed method shows significant potential for financial market prediction.
