Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Comprehensive Review: Applicability of Deep Neural Networks in Business Decision Making and Market Prediction Investment

Viet Trinh

TL;DR

This paper surveys how deep neural networks can support business decision making and market prediction in the context of big data. It reviews learning paradigms (supervised, unsupervised, reinforcement) and architectures (FNN, CNN, RNN), noting that in finance tasks inputs map from $\mathbb{R}^n$ to $\mathbb{R}^m$ and that ensemble approaches across modalities can boost performance. Key findings show deep networks improve prediction accuracy in risk management, portfolio optimization, and algorithmic trading, with evidence from CNN on image-like data, LSTM on time series, and meta-learners. It also identifies challenges—data privacy, real-time data availability, and cross-market transfer learning—and advocates future research on cross-modal networks and social-signal integration to build robust, scalable financial prediction frameworks.

Abstract

Big data, both in its structured and unstructured formats, have brought in unforeseen challenges in economics and business. How to organize, classify, and then analyze such data to obtain meaningful insights are the ever-going research topics for business leaders and academic researchers. This paper studies recent applications of deep neural networks in decision making in economical business and investment; especially in risk management, portfolio optimization, and algorithmic trading. Set aside limitation in data privacy and cross-market analysis, the article establishes that deep neural networks have performed remarkably in financial classification and prediction. Moreover, the study suggests that by compositing multiple neural networks, spanning different data type modalities, a more robust, efficient, and scalable financial prediction framework can be constructed.

A Comprehensive Review: Applicability of Deep Neural Networks in Business Decision Making and Market Prediction Investment

TL;DR

This paper surveys how deep neural networks can support business decision making and market prediction in the context of big data. It reviews learning paradigms (supervised, unsupervised, reinforcement) and architectures (FNN, CNN, RNN), noting that in finance tasks inputs map from to and that ensemble approaches across modalities can boost performance. Key findings show deep networks improve prediction accuracy in risk management, portfolio optimization, and algorithmic trading, with evidence from CNN on image-like data, LSTM on time series, and meta-learners. It also identifies challenges—data privacy, real-time data availability, and cross-market transfer learning—and advocates future research on cross-modal networks and social-signal integration to build robust, scalable financial prediction frameworks.

Abstract

Big data, both in its structured and unstructured formats, have brought in unforeseen challenges in economics and business. How to organize, classify, and then analyze such data to obtain meaningful insights are the ever-going research topics for business leaders and academic researchers. This paper studies recent applications of deep neural networks in decision making in economical business and investment; especially in risk management, portfolio optimization, and algorithmic trading. Set aside limitation in data privacy and cross-market analysis, the article establishes that deep neural networks have performed remarkably in financial classification and prediction. Moreover, the study suggests that by compositing multiple neural networks, spanning different data type modalities, a more robust, efficient, and scalable financial prediction framework can be constructed.

Paper Structure

This paper contains 12 sections, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Preliminary discussion on deep learning
  3. Learning paradigms
  4. Neural network models
  5. Financial applications
  6. Risk management
  7. Portfolio optimization
  8. Algorithmic trading
  9. Discussion
  10. Challenges
  11. Future research direction
  12. Conclusion

Figures (2)

  • Figure 1: Three paradigms in deep learning: supervised, unsupervised, and reinforcement. (a) In supervised learning, models are trained on labeled data to mostly perform the task of classification (where results are categories) and regression (where results are continuous values); (b) in unsupervised learning, models are trained on unlabeled data to identify associated relationships in a dataset, insightful cluster data group, and reduce high dimensional data form.
  • Figure 2: Neural network models: (a) Fully connected neural network; (b) Convolutional neural network; (c) Recurrent neural network (RNN).