Machine_Learning_Tidymodel_Linear.Rmd

---
title: "Machine_Learning_Tidymodel_Linear"
author: "Bernard Asante"
date: "2025-02-16"
output: 
  html_document:
    toc: true
    toc_float: true
    toc_depth: 3
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
library(tidyverse)
library(tidymodels)
library(themis)
library(parsnip)
library(gtsummary)
library(psych)
```

```{r}
data <- read_csv("Heart Attack.csv")

data %>% 
  glimpse()
```


```{r}
data <- data %>% 
  rename(pulse = impluse, sys = pressurehight, dias = pressurelow) %>% 
  mutate(glu_mmol = glucose/18)

data %>% 
  glimpse()
```


## Recoding variables 

```{r}
data <- data %>% 
  mutate(gender = case_when(
    gender == 1 ~ "Male",
    gender == 0 ~ "Female"
  )) %>% 
  mutate(gender = as.factor(gender)) %>% 
  mutate(class = as.factor(class))

data %>% 
  glimpse()
```


## Exploring missing data 

```{r}
missing_data <- data %>% 
  summarise_all(~sum(is.na(.)))

missing_data
```

# **Predictor Analysis**

## EDA of the predictor 

```{r}
ggplot(data, aes (troponin ))+
  geom_histogram()
```




```{r}
univ_table <- data %>% 
  tbl_uvregression(
    method = lm,
    y = troponin
  )

univ_table %>% as_kable()
```

## Correlation

### Correlation_matrix

```{r}

library(reshape2)
corr_matrix <- data %>% 
  select(age, pulse, glu_mmol,kcm,troponin) %>% 
  cor()

corr_matrix

melted_matrix <- melt(corr_matrix)

melted_matrix
```


```{r}
ggplot(melted_matrix, aes(x = Var1, y = Var2, fill = value))+
  geom_tile()+
  scale_fill_gradient2(low = "red", mid = "white", high = "blue", midpoint = 0)
```

```{r}
pairs.panels( data,
              scale = FALSE,
              method = "spearman",
              lm = TRUE,
              cor = TRUE,
              jiggle = FALSE,
              stars = FALSE)
```


# **Machine Leaning**

## Selecting variable 

```{r}
lin_data <- data %>% 
  select(age,gender, pulse,sys, dias,kcm,glu_mmol,troponin, class)

lin_data %>% 
  glimpse()


lin_data %>% 
  summary()
```

## Splitting Data

```{r}
set.seed(123)

data_split <- initial_split(lin_data, prop = 0.70)

train_data <- training(data_split)

test_data <- testing(data_split)
```


## Building Model 

```{r}
lin_ridge <- linear_reg(penalty= tune(), mixture = tune()) %>% 
  set_engine("glmnet")

```


## Building Recipe

```{r}
lin_recipe <- recipe(troponin ~ ., data = train_data) %>% 
  step_normalize(all_numeric_predictors()) %>% 
  step_dummy(all_nominal_predictors()) %>% 
  step_zv(all_predictors())
  
```


## Building workflow

```{r}
lin_workflow <- workflow() %>% 
  add_model(lin_ridge) %>% 
  add_recipe(lin_recipe)
```


## Hyperparameter Tunning


```{r}
fold = vfold_cv(train_data, v=5)

grid <- grid_regular(
  penalty(range = c(0.0001, 1)),  
  mixture(range = c(0, 1)),       # Mixture is between 0 (ridge) and 1 (lasso)
  levels = c(5, 5)                # Adjust the number of grid points
)


lin_tune_wrkflow <- tune_grid(lin_workflow,
                              resamples = fold,
                              grid = grid
)

lin_tune_wrkflow

```
## Selecting and Finalizing the best model


```{r}
best_params <- select_best(lin_tune_wrkflow, metric = "rmse")  # Change to "accuracy" for classification

final_wrkflow <- finalize_workflow(lin_workflow, best_params)
```


## Training the model 

```{r}
final_wrkflow_fit <- final_wrkflow %>% 
  fit(train_data)
```


```{r}
final_wrkflow_fit %>% 
  extract_fit_parsnip() %>% 
  tidy()
```




## Predictions 

```{r}
predict <- predict(final_wrkflow_fit, new_data = test_data)

predict %>% 
  head()
```

## Model Evaluation 

```{r}
model_eval <- bind_cols(test_data, predict)


model_eval %>% 
  glimpse()

```

## Accuracy 

```{r}
rmse <- rmse(model_eval, truth = troponin, estimate = .pred)

rmse

```