A LASSO model predicting an index tree’s linear distance from its nearest JBP-infested tree.
A Random-Forest Model predicting whether a tree is near another JBP-infested tree, i.e., within the 25% quantile of the linear distance from the nearest infested tree.
| True Class | Predicted Class | N | Result |
|---|---|---|---|
| Near JBP-Attacked Tree | Near JBP-Attacked Tree | 592 | True Positive |
| Near JBP-Attacked Tree | Distant from JBP-Attacked Tree | 22 | False Negative |
| Distant from JBP-Attacked Tree | Distant from JBP-Attacked Tree | 1808 | True Negative |
| Name | Piped data |
| Number of rows | 9687 |
| Number of columns | 18 |
| _______________________ | |
| Column type frequency: | |
| factor | 2 |
| numeric | 16 |
| ________________________ | |
| Group variables | None |
Variable type: factor
| skim_variable | n_missing | complete_rate | ordered | n_unique | top_counts |
|---|---|---|---|---|---|
| Response | 0 | 1 | FALSE | 2 | Ali: 8452, JPB: 1235 |
| NearBrood | 0 | 1 | FALSE | 2 | Dis: 7265, Nea: 2422 |
Variable type: numeric
| skim_variable | n_missing | complete_rate | mean | sd | p0 | p25 | p50 | p75 | p100 | hist |
|---|---|---|---|---|---|---|---|---|---|---|
| DeadDist | 0 | 1 | 24.08 | 18.84 | 0.11 | 9.52 | 19.84 | 34.72 | 124.95 | ▇▃▁▁▁ |
| TreeDiam | 0 | 1 | 13.11 | 6.05 | 3.00 | 9.00 | 12.00 | 16.00 | 78.00 | ▇▂▁▁▁ |
| Infest_Serv1 | 0 | 1 | 10.70 | 18.26 | 1.00 | 1.00 | 3.00 | 11.00 | 85.00 | ▇▁▁▁▁ |
| Infest_Serv2 | 0 | 1 | 10.93 | 21.47 | 0.14 | 0.55 | 1.58 | 10.65 | 92.40 | ▇▁▁▁▁ |
| SDI_20th | 0 | 1 | 26.35 | 12.72 | 0.33 | 17.20 | 25.59 | 34.45 | 92.86 | ▅▇▂▁▁ |
| Neigh_SDI_1/4th | 0 | 1 | 104.60 | 40.54 | 4.04 | 75.51 | 107.09 | 133.26 | 221.05 | ▂▅▇▅▁ |
| BA_20th | 0 | 1 | 16.54 | 8.01 | 0.14 | 10.97 | 16.00 | 21.43 | 58.38 | ▅▇▂▁▁ |
| Neigh_1/4th | 0 | 1 | 66.17 | 25.02 | 1.95 | 49.24 | 66.61 | 83.13 | 140.46 | ▂▅▇▃▁ |
| Neigh_1/2th | 0 | 1 | 122.79 | 40.98 | 4.54 | 97.55 | 125.02 | 150.25 | 230.42 | ▁▃▇▆▁ |
| Neigh_1 | 0 | 1 | 228.42 | 67.85 | 13.03 | 188.71 | 236.04 | 273.86 | 391.42 | ▁▃▇▇▂ |
| Neigh_1.5 | 0 | 1 | 328.16 | 93.54 | 22.24 | 270.60 | 338.88 | 393.72 | 541.62 | ▁▂▆▇▂ |
| BA_Inf_20th | 0 | 1 | 0.60 | 1.91 | 0.00 | 0.00 | 0.00 | 0.00 | 21.88 | ▇▁▁▁▁ |
| BA_Infest_1/4th | 0 | 1 | 2.69 | 6.04 | 0.00 | 0.00 | 0.00 | 2.63 | 72.30 | ▇▁▁▁▁ |
| BA_Infest_1/2th | 0 | 1 | 5.26 | 9.77 | 0.00 | 0.00 | 0.89 | 7.11 | 92.40 | ▇▁▁▁▁ |
| BA_Infest_1 | 0 | 1 | 10.30 | 16.49 | 0.00 | 0.27 | 3.13 | 12.78 | 106.67 | ▇▁▁▁▁ |
| BA_Infest_1.5 | 0 | 1 | 15.05 | 22.09 | 0.00 | 1.37 | 7.75 | 17.07 | 135.70 | ▇▁▁▁▁ |
---
title: "Predicting the Distance from a Jeffrey Pine Beetle-Infested Tree."
output:
flexdashboard::flex_dashboard:
orientation: columns
vertical_layout: fill
theme: sandstone
logo: JPBlogo.jpg
source_code: embed
---
```{r setup, include=FALSE}
library(tidyverse)
library(tidymodels)
library(readxl)
library(performance)
library(flexdashboard)
library(vip)
library(RColorBrewer)
library(patchwork)
library(ggthemes)
```
```{r data}
setwd("~/HGEN/2022-1 Spring/612 - Data Science II/hgen-612/assignments/project-1")
pine_dat <- read_excel("Data/Data_1993.xlsx", sheet = 1)
# Variable Transformation
pine_dat <- pine_dat %>%
mutate(Response = fct_recode(as_factor(Response),
"Alive" = "0", "JPB-attacked" = "1"),
NearBrood = ifelse(DeadDist <= quantile(DeadDist, probs = 0.25), 1, 0),
NearBrood = fct_recode(as_factor(NearBrood),
"Distant from JBP-Attacked Tree" = "0",
"Near JBP-Attacked Tree" = "1"),
NearBrood = fct_relevel(NearBrood, "Near JBP-Attacked Tree"))
# Variables
pine_tbl <- pine_dat %>%
select(Easting, Northing,
Response, DeadDist, NearBrood, TreeDiam, Infest_Serv1,
Infest_Serv2, SDI_20th, `Neigh_SDI_1/4th`, BA_20th,
`Neigh_1/4th`, `Neigh_1/2th`, Neigh_1, Neigh_1.5, BA_Inf_20th,
`BA_Infest_1/4th`, `BA_Infest_1/2th`, BA_Infest_1, BA_Infest_1.5)
```
LASSO Model {data-navmenu="Predicting the Distance"}
=====================================
Column {data-width=550}
-----------------------------------------------------------------------
### Clusters of Jeffrey Pine Beetle (JPB) Infestation in Lake Tahoe Basin in 1993
```{r jpb_plot, out.width='100%'}
ggplot(pine_tbl, aes(x = Easting, y = Northing)) +
geom_point(aes(color = Response, alpha = Response)) +
scale_alpha_discrete(range = c(0.25, 1), guide="none") +
scale_color_brewer("", palette = "Dark2") +
theme_minimal() + xlab("UTM X") + ylab("UTM Y")+
theme(legend.position = c(0,0.95), legend.text=element_text(size=10))
```
Column {data-width=450}
-----------------------------------------------------------------------
### Model Performance: R-squared ($R^2$)
A LASSO model predicting an index tree's linear distance from its nearest JBP-infested tree.
```{r lasso}
# Predicting the linear distance from the near infected tree
# Create training/testing data
set.seed(1234)
pine_split <- initial_split(pine_tbl)
pine_train <- training(pine_split)
pine_test <- testing(pine_split)
# bootstrap samples for resampling and tuning the penalty parameter
pine_boot <- bootstraps(pine_train)
# a grid of tuning parameters
lambda_grid <- grid_regular(penalty(), levels = 50)
lasso_mod <- linear_reg(mixture = 1, penalty = tune()) %>%
set_engine("glmnet")
# recipe
pine_rec <- pine_train %>%
recipe(DeadDist ~ TreeDiam + Infest_Serv1 + Infest_Serv2 + SDI_20th +
`Neigh_SDI_1/4th` + BA_20th + `Neigh_1/4th` + `Neigh_1/2th` +
Neigh_1 + Neigh_1.5 + BA_Inf_20th + `BA_Infest_1/4th` +
`BA_Infest_1/2th` + BA_Infest_1 + BA_Infest_1.5) %>%
step_corr(all_predictors()) %>%
step_normalize(all_numeric()) %>%
step_zv(all_numeric(), -all_outcomes())
# workflow
pine_lasso_wflow <-
workflow() %>%
add_model(lasso_mod) %>%
add_recipe(pine_rec)
# penalty parameter
set.seed(2020)
lasso_grid <- tune_grid(
pine_lasso_wflow,
resamples = pine_boot,
grid = lambda_grid
)
min_rmse <- lasso_grid %>%
select_best("rmse")
# update the final model with lowest rmse
final_lasso <- finalize_workflow(
pine_lasso_wflow,
min_rmse
)
# R-sq
lasso_rsq <- last_fit(
final_lasso,
pine_split
) %>%
collect_metrics() %>%
filter(.metric == "rsq") %>%
select(.estimate)
gauge(round(as.numeric(lasso_rsq*100),2), min = 0, max = 100, symbol = "%")
```
### Predictors Retained by the LASSO Model (out of 15 Variables)
```{r vip}
final_lasso %>%
fit(pine_train) %>%
extract_fit_parsnip() %>%
vi(lambda = min_rmse$penalty) %>%
mutate(
Importance = abs(Importance),
Variable = fct_reorder(Variable, Importance),
Sign = fct_relevel(Sign, "POS")
) %>%
ggplot(aes(x = Importance, y = Variable, fill = Sign)) +
geom_col() +
scale_x_continuous(expand = c(0, 0)) +
scale_fill_brewer(palette = "Dark2") +
labs(title = "", y = NULL) +
theme_minimal(15) +
theme(legend.position = c(1,0),
legend.justification = c(1,0))
```
Relationship with the Predictors {data-navmenu="Predicting the Distance"}
=====================================
Column {data-width=1000}
-----------------------------------------------------------------------
### Bivariate relationship with each of the Model-selected Predictors
```{r predictors, out.width='120%'}
pred1 <- ggplot(pine_tbl, aes(Neigh_1.5, DeadDist)) +
geom_smooth(method = "gam", color = "forestgreen") +
labs(title = "Basal Area of All Trees Within 1.5 Acre") +
theme_minimal(10) +
theme(plot.title.position = "plot",
plot.title = element_text(hjust = 0.5, size = 10),
axis.title = element_text(size = 8))
pred2 <- ggplot(pine_tbl, aes(BA_Infest_1.5, DeadDist)) +
geom_smooth(method = "gam", color = "forestgreen") +
labs(title = "Basal Area of Infested Trees Within 1.5 Acre") +
theme_minimal(10) +
theme(plot.title.position = "plot",
plot.title = element_text(hjust = 0.5, size = 10),
axis.title = element_text(size = 8))
pred3 <- ggplot(pine_tbl, aes(`BA_Infest_1/4th`, DeadDist)) +
geom_smooth(method = "gam", color = "forestgreen") +
labs(title = "Basal Area of Infested Trees Within 1/4th Acre") +
theme_minimal(10) +
theme(plot.title.position = "plot",
plot.title = element_text(hjust = 0.5, size = 10),
axis.title = element_text(size = 8))
pred4 <- ggplot(pine_tbl, aes(`Neigh_SDI_1/4th`, DeadDist)) +
geom_smooth(method = "gam", color = "forestgreen") +
labs(title = "Stand Density Index Within 1/4th Acre") +
theme_minimal(10) +
theme(plot.title.position = "plot",
plot.title = element_text(hjust = 0.5, size = 10),
axis.title = element_text(size = 8))
pred5 <- ggplot(pine_tbl, aes(BA_Inf_20th, DeadDist)) +
geom_smooth(method = "gam", color = "forestgreen") +
labs(title = "Basal Area of Infested Trees Within 1/20th Acre") +
theme_minimal(10) +
theme(plot.title.position = "plot",
plot.title = element_text(hjust = 0.5, size = 10),
axis.title = element_text(size = 8))
pred6 <- ggplot(pine_tbl, aes(Infest_Serv1, DeadDist)) +
geom_smooth(method = "gam", color = "forestgreen") +
labs(title = "Infestation Severity Nearest to Index Tree") +
theme_minimal(10) +
theme(plot.title.position = "plot",
plot.title = element_text(hjust = 0.5, size = 10),
axis.title = element_text(size = 8))
pred7 <- ggplot(pine_tbl, aes(BA_20th, DeadDist)) +
geom_smooth(method = "gam", color = "forestgreen") +
labs(title = "Basal Area of All Trees Within 1/20th Acre") +
theme_minimal(10) +
theme(plot.title.position = "plot",
plot.title = element_text(hjust = 0.5, size = 10),
axis.title = element_text(size = 8))
pred8 <- ggplot(pine_tbl, aes(TreeDiam, DeadDist)) +
geom_smooth(method = "gam", color = "forestgreen") +
labs(title = "Tree Diameter/Size") +
theme_minimal(10) +
theme(plot.title.position = "plot",
plot.title = element_text(hjust = 0.5, size = 10),
axis.title = element_text(size = 8))
(pred1 + pred2 + pred3 + pred4) / (pred5 + pred6 + pred7 + pred8)
```
Classifying by Distance
=====================================
Column {data-width=550}
-----------------------------------------------------------------------
### Clusters of Trees with Proximity to other Jeffrey Pine Beetle (JPB)-Infested Trees
```{r NearBrood, out.width='100%'}
ggplot(pine_tbl, aes(x = Easting, y = Northing)) +
geom_point(aes(color = NearBrood, alpha = NearBrood)) +
scale_alpha_discrete(range = c(0.5, 1), guide="none") +
scale_color_brewer("", palette = "Reds") +
labs(caption = "'Near' = Within the 25% Quantile of the Linear Distance\nFrom the Nearest JBP-Attacked Tree in this Sample.") +
theme_minimal() + xlab("UTM X") + ylab("UTM Y") +
theme(legend.position = c(0.1,0.95), legend.text=element_text(size=10))
```
Column {data-width=450}
-----------------------------------------------------------------------
### Model Performance: Area Under the ROC Curve (AUC)
A Random-Forest Model predicting whether a tree is near another JBP-infested tree, i.e., within the 25% quantile of the linear distance from the nearest infested tree.
```{r randfor}
# Predicting whether a tree is near or distant from an infested tree.
# Near = within 25% quantile of the distance from the nearest infested tree.
# recipe
dichot_recipe <- pine_train %>%
recipe(NearBrood ~ TreeDiam + Infest_Serv1 + Infest_Serv2 + SDI_20th +
`Neigh_SDI_1/4th` + BA_20th + `Neigh_1/4th` + `Neigh_1/2th` +
Neigh_1 + Neigh_1.5 + BA_Inf_20th + `BA_Infest_1/4th` +
`BA_Infest_1/2th` + BA_Infest_1 + BA_Infest_1.5) %>%
step_corr(all_predictors()) %>%
step_normalize(all_numeric()) %>%
step_zv(all_numeric(), -all_outcomes()) %>%
prep()
# model
rf_mod <- rand_forest(trees = 100, mode = "classification") %>%
set_engine("ranger")
# workflow
pine_rf_wflow <-
workflow() %>%
add_model(rf_mod) %>%
add_recipe(dichot_recipe)
# train
pine_rf <- pine_rf_wflow %>%
fit(pine_test)
# test
pine_rf_auc <- predict(pine_rf, pine_test, type = "prob") %>%
bind_cols(predict(pine_rf, pine_test)) %>%
bind_cols(select(pine_test, NearBrood)) %>%
metrics(truth = NearBrood, `.pred_Near JBP-Attacked Tree`,
estimate = .pred_class) %>%
filter(.metric == "roc_auc") %>%
select(.estimate)
gauge(round(as.numeric(pine_rf_auc*100),3), min = 0, max = 100, symbol = "%")
```
### Model Performance: Accuracy
```{r accuracy}
pine_rf_acc <- pine_rf %>%
predict(pine_test) %>%
bind_cols(pine_test) %>%
metrics(truth = NearBrood, estimate = .pred_class) %>%
filter(.metric == "accuracy") %>%
select(.estimate)
gauge(round(as.numeric(pine_rf_acc*100),3), min = 0, max = 100, symbol = "%")
```
### Classification by the Random-Forest Model
- ***No False Positives.***
```{r RFtable}
predict(pine_rf, pine_test) %>%
bind_cols(pine_test) %>%
count(NearBrood, .pred_class) %>%
rename(`True Class` = NearBrood,
`Predicted Class` = .pred_class,
N = n) %>%
mutate(Result = c("True Positive","False Negative","True Negative")) %>%
knitr::kable(align = 'llrr')
```
Data Overview
=====================================
### Variables Used in the Models
```{r skim, out.width='100%'}
pine_tbl %>%
select(-c(Easting, Northing)) %>%
skimr::skim()
```