claims_EDA.Rmd

---
title: "French Motor Claims EDA"
author: "Xiaoxi"
date: "03/03/2021"
output: html_document
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```

## French motor claims EDA

### Load libraries

```{r}
library(tidyverse)
library(vroom)
```

### Load dataset

```{r}
dataset <- vroom('freMTPL2freq.csv')
```

```{r}
str(dataset)
```

```{r}
dataset %>% 
  head()
```

### EDA

#### Exposure

```{r}
dataset %>%
  ggplot(aes(x = Exposure)) +
  geom_histogram(binwidth=0.1) +
  labs(title = 'histogram of the exposures',
       x = 'exposures',
       y = 'number of policies')
```

```{r}
dataset %>% 
  ggplot(aes(y=Exposure)) +
  geom_boxplot() +
  labs(title = 'boxplot of the exposures', y = '')
```

```{r}
dataset %>% 
  ggplot(aes(x = factor(ClaimNb))) +
  geom_bar() +
  labs(title = 'histogram of claim numbers',
       x = 'number of claims',
       y = 'frequency')
```

```{r}
dataset %>%
  group_by(ClaimNb) %>% 
  summarise(number_policies = n(),
            total_exposure = sum(Exposure))
```

#### Data transformation

```{r}
dataset <- 
  dataset %>% 
  mutate(ClaimNb = pmin(ClaimNb, 4), # correct for unreasonable observations
         Exposure = pmin(Exposure, 1), # correct for unreasonable observations
         VehAge = pmin(VehAge, 20),
         DrivAge = pmin(DrivAge, 90),
         BonusMalus = pmin(round(BonusMalus, -1), 150))
```

```{r}
dataset %>%
  group_by(ClaimNb) %>% 
  summarise(number_policies = n(),
            total_exposure = sum(Exposure))
```

```{r}
dataset %>% 
  summarise(freq = 100 * sum(ClaimNb) / sum(Exposure))
```

##### Auxiliar functions

```{r}
hist_expo <- function(dataset, var_group, var_text) {
  dataset %>% 
  ggplot(aes(x = {{var_group}}, y = Exposure)) +
  geom_col() +
  labs(title = paste('total volumes per', var_text),
       x = var_text,
       y = 'exposure')
}
```

```{r}
plot_freq <- function(dataset, var_group, var_text, ymax) {
  dataset %>% 
  group_by({{var_group}}) %>% 
  summarise(total_exp = sum(Exposure),
    freq_mean = sum(ClaimNb) / sum(Exposure)) %>% 
  mutate(std = sqrt(freq_mean / total_exp),
         freq_min = freq_mean - 2*std,
         freq_max = freq_mean + 2*std) %>%
  select({{var_group}}, starts_with('freq')) %>%
  mutate(across(starts_with('freq'), ~round(.x, 4))) %>%
  ggplot() +
    geom_point(aes(x = {{var_group}}, y = freq_mean), shape = 19) +
    stat_summary(aes(x = {{var_group}}, y = freq_max, group = 1), 
                 fun = sum, geom = 'line', lty = 'dashed', col = 'blue') +
    stat_summary(aes(x = {{var_group}}, y = freq_min, group = 1), 
                 fun = sum, geom = 'line', lty = 'dashed', col = 'blue') +
    labs(title = paste('observed frequency per', var_text),
         x = var_text,
         y = 'frequency') +
    ylim(c(0, ymax))
}
```

#### Area Code

```{r}
hist_expo(dataset, Area, 'area code groups')
```

```{r}
plot_freq(dataset, Area, 'area code groups', 0.35)
```

#### Vehicle power

```{r}
hist_expo(dataset, factor(VehPower), 'vehicle power groups')
```

```{r}
summary_freq(dataset, VehPower)
```

```{r}
plot_freq(dataset, VehPower, 'vehicle power groups', 0.35)
```

#### Vehicle age

```{r}
hist_expo(dataset, VehAge, 'vehicle age groups')
```

```{r}
plot_freq(dataset, VehAge, 'vehicle age groups', 0.35)
```

#### Driver's age

```{r}
hist_expo(dataset, DrivAge, 'driver age groups')
```

```{r}
plot_freq(dataset, DrivAge, 'driver age groups', 0.35)
```

#### Bonus-malus

```{r}
hist_expo(dataset, factor(BonusMalus), 'bonus-malus level groups')
```

```{r}
plot_freq(dataset, BonusMalus, 'bonus-malus groups', 0.7)
```

#### Car brand

```{r}
hist_expo(dataset, VehBrand, 'car brand groups')
```

```{r}
plot_freq(dataset, VehBrand, 'car brand groups', 0.35)
```

#### Fuel type

```{r}
hist_expo(dataset, VehGas, 'fuel type')
```

```{r}
plot_freq(dataset, VehGas, 'fuel type', 0.35)
```

#### Density

```{r}
dataset %>% 
  mutate(Density = round(log(Density))) %>%
  hist_expo(Density, 'density (log-scale) groups')
```

```{r}
dataset %>%
  mutate(Density = round(log(Density))) %>% 
  plot_freq(Density, 'density (log-scale) groups', 0.35)
```

#### Region

```{r}
hist_expo(dataset, VehAge, 'regional groups')
```

```{r}
plot_freq(dataset, Region, 'regional groups', 0.35)
```