FeatureSaliency Hidden Markov Model

This repository presents an implementation of the feature saliency HMM algorithm as proposed by Adams et. al in the paper Feature Selection for Hidden Markov Models and Hidden Semi-Markov Models. The implementation is a modification of the GaussianHMM class of hmmlearn.

Requirements:

Install dependencies via pip install -r requirements.txt

hmmlearn==0.2.0
scikit-learn==0.19.1

Quick Start:

The notebook FSHMM_example.ipynb has a short example on how to use the library and shows a simple test case.

Resources:

• Feature Selection for Hidden Markov Models and Hidden Semi-Markov Models: Original paper where FSHMM is presented.
• A novel dynamic asset allocation system using Feature Saliency Hidden Markov models for smart beta investing: Paper that uses FSHMM for regime identification in financial markets.
• hmmlearn: python library for HMMs.
• Simultaneous Feature Selection and Parameter Estimation for Hidden Markov Models: PhD thesis with detailed derivation of the algorithm.

If you use this code, please cite the following papers:

@article{fons2021,
	title = {A novel dynamic asset allocation system using Feature Saliency Hidden Markov models for smart beta investing},
	author = {Elizabeth Fons and Paula Dawson and Jeffrey Yau and Xiao-jun Zeng and John Keane},
	journal = {Expert Systems with Applications},
	volume = {163},
	year = {2021},
	pages = {113720},
	issn = {0957-4174},
	doi = {10.1016/j.eswa.2020.113720}
}

@article{adams2016,  
	title={Feature Selection for Hidden Markov Models and Hidden Semi-Markov Models},   
	author={Stephen Adams and Peter A. Beling and Randy Cogill},  
	journal={IEEE Access},   
	volume={4},  
	year={2016},  
	pages={1642-1657},  
	doi={10.1109/ACCESS.2016.2552478}
}