Insurance pricing game

Insurance pricing game: EDA

Hello everyone! The notebook will be more useful for new competitors. It will help you save time analyzing your data and get a more detailed overview. Hope this is helpful) Also, if I find it useful, I will continue to supplement the notebook.

Link to notebook in Kaggle

Support like it if it is helpful

This notebook will examine the competition data in more detail! Ideal for new members.

In [1]:

import numpy as np
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
import random

In [2]:

df = pd.read_csv('/kaggle/input/motor-insurance-market/training (1).csv')

In [3]:

df.head()

Out[3]:

	id_policy	year	pol_no_claims_discount	pol_coverage	pol_duration	pol_sit_duration	pol_pay_freq	pol_payd	pol_usage	drv_sex1	...	vh_make_model	vh_age	vh_fuel	vh_type	vh_speed	vh_value	vh_weight	population	town_surface_area
0	PL000000	1.0	0.332	Med2	5	1	Monthly	No	WorkPrivate	M	...	aparvvfowrjncdhp	8.0	Gasoline	Tourism	174.0	11040.0	1143.0	1270.0	33.1
1	PL042495	1.0	0.000	Med2	6	1	Monthly	No	WorkPrivate	M	...	aparvvfowrjncdhp	10.0	Diesel	Tourism	174.0	11040.0	1143.0	1290.0	51.3
2	PL042496	1.0	0.196	Med1	2	1	Yearly	Yes	Retired	M	...	iwhqpdfuhrsxyqxe	8.0	Diesel	Commercial	150.0	14159.0	1193.0	1020.0	262.8
3	PL042497	1.0	0.000	Med2	8	5	Yearly	No	WorkPrivate	F	...	kvcddisqpkysmvvo	4.0	Gasoline	Tourism	149.0	17233.0	1012.0	180.0	219.7
4	PL042498	1.0	0.000	Med1	2	2	Yearly	No	Retired	F	...	tdgkjlphosocwbgu	13.0	Gasoline	Tourism	200.0	19422.0	1315.0	30.0	70.3

5 rows × 26 columns

Understanding data¶

The training set presents policy data for the last 4 years for each user. Let's check it out.

In [4]:

df.id_policy.value_counts()

Out[4]:

PL008115    4
PL011513    4
PL057669    4
PL033195    4
PL043079    4
           ..
PL088572    4
PL052667    4
PL036222    4
PL028661    4
PL006028    4
Name: id_policy, Length: 57054, dtype: int64

Let's get acquainted with ordinal, categorical and binary features:¶

Ordinal:

pol_coverage: Min, Med1, Med2, Max, in this order.

Categorical:

pol_pay_freq: the price of the insurance coverage can be paid anually, bi-anually, quarterly or monthly.
pol_usage: WorkPrivate, Retired, Professional, AllTrips.
drv_sex1: H or F.
drv_sex2: H or F.
vh_make_model: hashes representing car models, such as Honda Civic or Fiat Uno.
vh_fuel: Diesel, Gasoline or Hybrid.
vh_type: Tourism or Commercial.

Boolean:

pol_payd: Pay As You Drive. Indicates whether a client subscribed a mileaged-base policy or not.
drv_drv2: Indicates the presence of a secondary driver in the contract.

Missing data¶

In [5]:

df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 228216 entries, 0 to 228215
Data columns (total 26 columns):
 #   Column                  Non-Null Count   Dtype  
---  ------                  --------------   -----  
 0   id_policy               228216 non-null  object 
 1   year                    228216 non-null  float64
 2   pol_no_claims_discount  228216 non-null  float64
 3   pol_coverage            228216 non-null  object 
 4   pol_duration            228216 non-null  int64  
 5   pol_sit_duration        228216 non-null  int64  
 6   pol_pay_freq            228216 non-null  object 
 7   pol_payd                228216 non-null  object 
 8   pol_usage               228216 non-null  object 
 9   drv_sex1                228216 non-null  object 
 10  drv_age1                228216 non-null  float64
 11  drv_age_lic1            228216 non-null  float64
 12  drv_drv2                228216 non-null  object 
 13  drv_sex2                228216 non-null  object 
 14  drv_age2                75320 non-null   float64
 15  drv_age_lic2            75320 non-null   float64
 16  vh_make_model           228216 non-null  object 
 17  vh_age                  228212 non-null  float64
 18  vh_fuel                 228216 non-null  object 
 19  vh_type                 228216 non-null  object 
 20  vh_speed                225664 non-null  float64
 21  vh_value                225664 non-null  float64
 22  vh_weight               225664 non-null  float64
 23  population              228216 non-null  float64
 24  town_surface_area       228216 non-null  float64
 25  claim_amount            228216 non-null  float64
dtypes: float64(13), int64(2), object(11)
memory usage: 45.3+ MB

Note that there is a lot of missing data associated with the second driver. This is most likely due to the fact that there is no second driver. These missing values may need to be categorized into new categories.

Data analysis¶

Correlations¶

Unfortunately, there are very few dependencies on the target variable. Getting good predictions won't be easy.)

In [6]:

sns.heatmap(df.corr())

Out[6]:

<matplotlib.axes._subplots.AxesSubplot at 0x7f508cc4fc50>

Target¶

Let's see what the distribution of the target variable looks like. Expectedly, there are many zeros, which means that the person was not hurt. There is also a long right tail.

In [7]:

sns.distplot(df['claim_amount'], kde=False)

Out[7]:

<matplotlib.axes._subplots.AxesSubplot at 0x7f508c9a2250>

In [8]:

sns.distplot(df[(df['claim_amount'] > 0) & (df['claim_amount'] < 10000)]['claim_amount'], kde=False)

Out[8]:

<matplotlib.axes._subplots.AxesSubplot at 0x7f508c817e90>

In [9]:

sns.boxplot(df['claim_amount'])

Out[9]:

<matplotlib.axes._subplots.AxesSubplot at 0x7f508c70d050>

Now let's look at the ratio of those who have suffered and not.

In [10]:

df['Damage'] = np.where(df['claim_amount'], 1, 0)
sns.countplot(df['Damage'])

Out[10]:

<matplotlib.axes._subplots.AxesSubplot at 0x7f508c71b250>

Categorical features¶

In [11]:

sns.countplot('pol_coverage', data=df, hue = 'Damage')

Out[11]:

<matplotlib.axes._subplots.AxesSubplot at 0x7f5087759ad0>

In [12]:

sns.countplot('pol_pay_freq', data=df, hue ='Damage')

Out[12]:

<matplotlib.axes._subplots.AxesSubplot at 0x7f5086339a90>

In [13]:

sns.countplot('pol_usage', data=df, hue='Damage')

Out[13]:

<matplotlib.axes._subplots.AxesSubplot at 0x7f50862a5a50>

In [14]:

sns.countplot('drv_sex1', data=df, hue='Damage')

Out[14]:

<matplotlib.axes._subplots.AxesSubplot at 0x7f5086238f50>

In [15]:

sns.countplot('vh_fuel',data=df, hue='Damage')

Out[15]:

<matplotlib.axes._subplots.AxesSubplot at 0x7f50861a3950>

In [16]:

sns.countplot('vh_type',data=df, hue='Damage')

Out[16]:

<matplotlib.axes._subplots.AxesSubplot at 0x7f5086122590>

Crashes¶

Find out how many users have not received damage for 4 years, as well as the rest who have received damage.

In [17]:

count_damage_per_user = df.groupby('id_policy')['Damage'].sum()
count_damage_per_user.value_counts()

Out[17]:

0    38518
1    14440
2     3495
3      542
4       59
Name: Damage, dtype: int64

Let's explore which data depends and changes over time. For some users, they received that their gender was changed. I think we need to investigate this in more detail.

In [18]:

no_change, change = [], []
for column in df.columns:
    constant_cells = (df.groupby(['id_policy'])[column].value_counts() == 4).all()
    if constant_cells:
        no_change.append(column)
    else:
        change.append(column)

In [19]:

change

Out[19]:

['year',
 'pol_no_claims_discount',
 'pol_duration',
 'pol_sit_duration',
 'pol_pay_freq',
 'drv_sex1',
 'drv_age1',
 'drv_age_lic1',
 'drv_age2',
 'drv_age_lic2',
 'vh_age',
 'population',
 'town_surface_area',
 'claim_amount',
 'Damage']

Content

1934

Show Comments

Comments

MakePredict

Almost 4 years ago

I read it. Not all claims are due to damage to the insured vehicle, if you read the description of pol_coverage you’ll see that all categories cover third party liability.

Liked by

mrmorj

Almost 4 years ago

Thank you for your comment!

You must login before you can post a comment.