WhoseEgg

.title[
# WhoseEgg
]
.subtitle[
## .white[A Shiny App for Identifying Invasive Carp Using Random Forests and Fish Egg Characteristics]
]
.author[
### <br>
]
.author[
### .white[Katherine Goode]
]
.author[
### .white[Joint work with] <a href="https://michaeljohnweber.wixsite.com/fisherieslab">Dr. Mike Weber</a> .white[and] <a href="https://pdixon.stat.iastate.edu/">Dr. Philip Dixon</a>
]
.author[
### .white[Presented at ISU Graphics Group]
]
.date[
### .white[February 25, 2021] <br> <br> <br> <br> .medium[.white[Code for slide generation found <a href="https://github.com/goodekat/presentations/tree/master/2021-isugg-shiny-fish">here</a>]]
]

---

# Overview

- Background on invasive carp
  
<br>

- Random forest approach to fish egg identification

<br>

- Shiny app

]

# Background on Invasive Carp

---

# Invasive Carp

.pull-left[
Information from [National Park Service](https://www.nps.gov/miss/learn/nature/ascarpover.htm):

- Native to Europe and Asia

- Brought to US in 1970's "to control algae, weed, and parasite growth in aquatic farms, weeds in canal systems, and as one form of sewage treatment"

- Escaped into wild

- Become a problem in Mississippi River basin:

- Threaten native fish populations
  - May lower water quality
  - Possible danger to boaters
  
]

---

# Jumping Carp

---

# More Jumping Carp

---

# Management of Carp

- Federal and state agencies monitor and try to manage invasive carp populations to:

- Control population
  - Keep out of new aquatic ecosystems (include Great Lakes)

- Management practices from [USGS](https://www.usgs.gov/ecosystems/invasive-species-program/science/asian-carp?qt-science_center_objects=0#qt-science_center_objects)

---

# Egg Collection

<br>

- One method for monitoring carp is to collect eggs

- Common practice is to have eggs identified using genetic analysis

- Both expensive and timely

- Led Camacho et al (2019) to try using machine learning to classify fish eggs

<br>

.center[
<img src="figures/egg1.png" width="30%" /><img src="figures/egg2.png" width="30%" /><img src="figures/egg5.png" width="30%" />
]

---

# Random Forest Approach to Egg Classification

---

# Overview of Random Forests

.pull-left[
**[Camacho et al. (2019)](https://afspubs.onlinelibrary.wiley.com/doi/abs/10.1002/nafm.10380)**

- Trained random forests

- Predictor variables: egg characteristics

- Responses: taxonomy levels (family, genus, species)

- Models showed promise for identifying invasive carp

]

**[Goode et al. (submitted manuscript)](https://github.com/goodekat/carp-egg-rf-validation)**

- Validated models using a new year of data

- Trained new models with all years of data

<br>

![](figures/large-carp2.JPG)
]

---

# Sampling Site and Years

Three years of collection

---

# Egg Collection

Collected eggs were taken to the lab for measurements and genetic identification

---

# Predictor Variables

- Month
- Julian day  
- Water temperature	 
- Water conductivity  	
- Deflated?	 
- Pigment?  
- Compact or diffuse?  	
- Sticky debris?  
- Egg stage  
- Membrane average  
- Membrane standard deviation  
- Membrane coefficient of variation  
- Yolk average
- Yolk standard deviation  
- Yolk coefficient of variation
- Yolk to membrane ratio
- Larval length  
]

<img src="figures/egg-characteristics.png" width="100%" style="display: block; margin: auto;" />
]

---

# Response Variables

**Taxonomy**:

- Genus (17 levels)

- Species (29 levels)

<br>

**Taxonomy with invasive carp categorized as on class**:

- Family (8 levels)

- Genus (16 levels)

- Species (27 levels)

]

<br> 
<small> Image source: https://www.watershedcouncil.org/asian-carp.html </small>
]

---

# Results: Metrics on Training Data

Models trained with 2014-2015 and 2014-2016 data

---

# Results: Validation of Original Models

Metrics computed using 2014-2015 model on 2016 data

---

# Shiny App

---

# WhoseEgg

**Goal**

- Provide a tool for easily predicting the identification of fish eggs using the random forests

- Used to help identify areas where invasive carp are reproducing

<br>

**Audience**

- Scientists and natural resource management organization employees

- Many non-R users

- Users in upper Mississippi River basin (possibly other regions)

<br>

**Models used**

- Random forests for predicting family, genus, and species with invasive carp grouped

---

# Feedback from Audience

**Current version:**

- Version 1 of app (almost complete)

- Have ideas for improvement

- Would be great to hear thoughts!

<br>

**For example:**

- Suggestions to make more user friendly

- Data input process?

- What else to look at with predictions within the app?

- Advice for instructions?

- Organizing help page?

---

# App Demonstration

---

# Plans for Adding to the App

Version 2:

- Add visualizations of input data compared to training data

- Include prediction intervals

<br>

Version 3:

- Add visualizations relating to explainability for individual predictions

- Add video showing how to use the app

- Add image during loading of swimming fish

---

# Thank you!

---

# Additional Figures

---

# Metrics on New/Old Sites

---

# Feature Importance

---

# Partial Dependence Plots

---

# Correlation Between Features