MCN portfolio: An efficient portfolio prediction and selection model using multiserial cascaded network with hybrid...

Area of Science:

• Financial modeling and computational intelligence.
• Data science and machine learning applications in finance.

Background:

• Portfolio management requires accurate financial investment predictions, which are often complicated by existing techniques.
• Effective error analysis and performance measures are crucial for validating portfolio prediction models.

Purpose of the Study:

• To develop a novel framework for portfolio prediction and optimization.
• To enhance the accuracy of financial forecasting and identify optimal investment portfolios.
• To address the complexities and issues in current portfolio prediction methods.

Main Methods:

• A Multi-serial Cascaded Network (MCNet) was employed, integrating Autoencoder, 1D Convolutional Neural Network (1DCNN), and Recurrent Neural Network (RNN) for benefit forecasting.
• A dataset of company portfolios was collected for training and validation.
• The Integration of Artificial Rabbit and Hummingbird Algorithm (IARHA) was used to select the optimal portfolio based on predicted profits.

Main Results:

• The MCNet model demonstrated strong performance in forecasting company benefits.
• The framework achieved low error rates, with Root Mean Square Error (RMSE) at 0.89% and Mean Absolute Error (MAE) at 0.56%.
• The developed model showed enriched performance in experimental analysis.

Conclusions:

• The proposed portfolio prediction framework significantly increases prediction accuracy.
• The integration of MCNet and IARHA effectively aids in selecting optimal investment portfolios.
• The study highlights the potential of advanced machine learning techniques for improved financial decision-making.