MAP-FCRNN: Multi-step ahead prediction model using forecasting correction and RNN model with memory functions

Information Sciences, Год журнала: 2023, Номер 646, С. 119382 - 119382

Опубликована: Июль 18, 2023

Язык: Английский

---

An enhanced interval-valued decomposition integration model for stock price prediction based on comprehensive feature extraction and optimized deep learning

Expert Systems with Applications, Год журнала: 2023, Номер 243, С. 122891 - 122891

Опубликована: Дек. 12, 2023

Язык: Английский

---

A hybrid neural network and optimization algorithm for forecasting and trend detection of Forex market indices

Decision Analytics Journal, Год журнала: 2023, Номер 6, С. 100193 - 100193

Опубликована: Март 1, 2023

An Autoencoder (AE) is an independent feature extractor from data samples and a deep network can be obtained by stacking several AEs. This paper presents novel hybrid stacked Autoencoder-based Deep Kernel-based Random Vector Functional Link Network (DKRVFLN-AE) for forecasting trend analysis of Foreign Exchange (Forex) rates. The proposed model dispenses the random choices weights biases, unlike Network-AE (DRVFLN-AE), using wavelet kernel function with strong fitting capability based on Mercer's condition. A modified metaheuristic Water Cycle Algorithm used to optimize parameters provide DKRVFLN-AE better generalization learning capability, faster execution speed, lower storage space, improved accuracy traditional Extreme Learning Machine models. Applications this new approach predict exchange rates three foreign markets successful results validate its superiority over well-known approaches like Networks, Support Machines, Naive-Bayes, Network.

Язык: Английский

---

Multi-factor stock trading strategy based on DQN with multi-BiGRU and multi-head ProbSparse self-attention

Applied Intelligence, Год журнала: 2024, Номер 54(7), С. 5417 - 5440

Опубликована: Апрель 1, 2024

Abstract Reinforcement learning is widely used in financial markets to assist investors developing trading strategies. However, most existing models primarily focus on simple volume-price factors, and there a need for further improvement the returns of stock trading. To address these challenges, multi-factor strategy based Deep Q-Network (DQN) with Multi-layer Bidirectional Gated Recurrent Unit (Multi-BiGRU) multi-head ProbSparse self-attention proposed. Our comprehensively characterizes determinants prices by considering various factors such as quality, valuation, sentiment factors. We first use Light Gradient Boosting Machine (LightGBM) classify turning points data. Then, reinforcement strategy, Multi-BiGRU, which holds bidirectional historical data, integrated into DQN, aiming enhance model’s ability understand dynamics market. Moreover, mechanism effectively captures interactions between different providing model deeper market insights. validate our strategy’s effectiveness through extensive experimental research stocks from Chinese US markets. The results show that method outperforms both temporal non-temporal terms returns. Ablation studies confirm critical role LightGBM mechanism. experiment section also demonstrates significant advantages presentation box plots statistical tests. Overall, fully data feature extraction capabilities, work expected provide more precise decision support. Graphical abstract

Язык: Английский

---

Multi-step-ahead copper price forecasting using a two-phase architecture based on an improved LSTM with novel input strategy and error correction

Resources Policy, Год журнала: 2022, Номер 79, С. 102962 - 102962

Опубликована: Сен. 10, 2022

Язык: Английский

---