Contributed by: Manorama Yadav
LinkedIn Profile: https://www.linkedin.com/in/manorama-3110/
Introduction to EDA in Python
Exploratory data analysis is the analysis of the data and brings out the insights. It’s storytelling, a story which data is trying to tell. EDA is an approach to analyse the data with the help of various tools and graphical techniques like barplot, histogram etc.
According to Tukey (data analysis in 1961)
“Procedures for analysing data, techniques for interpreting the results of such procedures, ways of planning the gathering of data to make its analysis easier, more precise or more accurate, and all the machinery and results of (mathematical) statistics which apply to analysing data.”
EDA in Python
There are many libraries available in python like pandas, NumPy, matplotlib, seaborn etc. with the help of those we can do the analysis of the data and bring out helpful insights. I will be using Jupyter Notebook along with these libraries.
The dataset I am using is the ‘Cars’ dataset which has different features of cars like model, year, engine and other properties along with its price. It has 28 years of data from 1990 to 2017. You can download the dataset here.
|4||Engine Fuel Type||Fuel Type||String|
|5||Engine HP||Horse Power(HP)||Integer|
|6||Engine Cylinders||No. of Cylinders||Integer|
|7||Transmission Type||Transmission Type||String|
|9||Number of Doors||No. of Doors||Integer|
|10||Market Category||Market Category||String|
|11||Vehicle Size||Size of Vehicle||String|
|12||Vehicle Style||Type of Vehicle||String|
|13||highway MPG||Highway MPG||Integer|
|14||city mpg||miles per gallon||Integer|
|15||Popularity||Popularity of the car||Integer|
|16||MSRP||Price of the car in ($)||Integer|
The objective of this article is to explore the data and make it ready for modelling.
Let’s get started!!!
Exploratory Data Analysis in Python
First of all, we will import all the libraries that are required for EDA (Exploratory Data Analysis). This is the first and most important thing to do. Without importing libraries we will not be able to perform anything.
After importing the libraries, the next step is loading data into the dataframe. To load the data into the dataframe, we will use pandas library. It supports various file formats like Comma Separated Values (.csv), excel (.xlsx, .xls) etc.
To read the dataset, either store the data file into the same directory and read it directly, or provide the path of the data file where the dataset is located while reading the data.
Top 5 rows
Now, the data has been loaded. Let’s check the first 5 rows of the dataset.
From the above results, we can see that the index in python starts from 0.
Bottom 5 rows
To check the dimensions of the dataframe, let’s check the number of rows and columns present in the dataset.
Shape of the Data
There are a total of 11914 rows and 16 columns in the dataset
Concise info of dataset
Now, check the data types along with the concise summary of all the variables in the dataset. It includes the number of non-null values present.
<class 'pandas.core.frame.DataFrame'> RangeIndex: 11914 entries, 0 to 11913 Data columns (total 16 columns): Make 11914 non-null object Model 11914 non-null object Year 11914 non-null int64 Engine Fuel Type 11911 non-null object Engine HP 11845 non-null float64 Engine Cylinders 11884 non-null float64 Transmission Type 11914 non-null object Driven_Wheels 11914 non-null object Number of Doors 11908 non-null float64 Market Category 8172 non-null object Vehicle Size 11914 non-null object Vehicle Style 11914 non-null object highway MPG 11914 non-null int64 city mpg 11914 non-null int64 Popularity 11914 non-null int64 MSRP 11914 non-null int64 dtypes: float64(3), int64(5), object(8) memory usage: 1.5+ MB
The type of data will be stored as an object if there are strings present in the variables. Also, it will be int or float if the data has numerical and decimal values respectively. MSRP (the price of the car) is stored as int data type while Driven_wheels is stored as an object data type.
Above results show many variables like Engine Fuel Type, Engine HP, Engine Cylinders, No. of Doors, and Market Category have missing values in the data.
We can check the data types by one more method:
Make object Model object Year int64 Engine Fuel Type object Engine HP float64 Engine Cylinders float64 Transmission Type object Driven_Wheels object Number of Doors float64 Market Category object Vehicle Size object Vehicle Style object highway MPG int64 city mpg int64 Popularity int64 MSRP int64 dtype: object
To print the columns of the dataset
Index(['Make', 'Model', 'Year', 'Engine Fuel Type', 'Engine HP', 'Engine Cylinders', 'Transmission Type', 'Driven_Wheels', 'Number of Doors', 'Market Category', 'Vehicle Size', 'Vehicle Style’, ‘highway MPG', 'city mpg', 'Popularity', 'MSRP'], dtype='object')
Since the names of the columns are very lengthy, let’s rename them.
Rename the Columns
Drop the columns which are not necessary for the dataframe. It is not necessary that all the columns in the data are relevant. In this data, columns like popularity, number of doors, vehicle_size was not so relevant. So I am dropping these variables from the dataset.
Make 0 Model 0 Year 0 Fuel_Type 3 HP 69 Cylinders 30 Transmission 0 Driven_Wheels 0 Vehicle_Style 0 h_mpg 0 c_mpg 0 price 0 dtype: int64
Above results show that out of 12 variables, 3 variables Fuel_type, HP, and cylinders have missing values.
Let’s check the percentage of the data are missing column wise
Make 0.000000 Model 0.000000 Year 0.000000 Fuel_Type 0.025180 HP 0.579151 Cylinders 0.251805 Transmission 0.000000 Driven_Wheels 0.000000 Vehicle_Style 0.000000 h_mpg 0.000000 c_mpg 0.000000 price 0.000000 dtype: float64
There are 0.025%, 0.58% and 0.25% data are missing in the variables Fuel_type, HP and cylinders respectively.
There are many ways to treat these missing values.
We can either drop the rows where missing values are present or replace the missing values with some values like mean, median or mode.
Since the % of the data missing is very less, we can remove those rows from the dataset.
Make 0 Model 0 Year 0 Fuel_Type 0 HP 0 Cylinders 0 Transmission 0 Driven_Wheels 0 Vehicle_Style 0 h_mpg 0 c_mpg 0 price 0 dtype: int64
By default, the drop function will drop the complete row if any of the variables has missing values.
After dropping the missing values, now the count of missing values is 0. That means there are no missing values present in the dataset.
Check the number of rows present after removing the missing values.
Make 11813 Model 11813 Year 11813 Fuel_Type 11813 HP 11813 Cylinders 11813 Transmission 11813 Driven_Wheels 11813 Vehicle_Style 11813 h_mpg 11813 c_mpg 11813 price 11813 dtype: int64
The original number of rows was 11914, now the number of rows left is 11813.
Now, let’s find out the statistical summary or 5-point summary of the dataset. The 5-point summary tells the descriptive summary which includes mean, median, mode, no. of rows, maximum value, and minimum value for each variable.
Mean, standard deviation, max, and percentile values will be NaN for variables which have object datatype.
The unique, top, frequency will be NaN for variables which have int data type.
From the descriptive summary, we got to know that there is 47 unique make of the cars and 904 models. Data has maximum Chevrolet make cars with 1115 counts. The average price of the car is 40581.5 dollars. The 50th percentile or median of the price is 29970. There is a huge difference between the mean and median of the price. This depicts that the price variable is highly skewed, which we can check visually using a histogram.
Data visualisation, as its name suggests, is to observe the data using various types of plots, graphs etc. Various plots include histogram, scatterplot, boxplot, heatmap etc. We will use matplotlib and seaborn together to visualise a few variables.
Histogram (Distribution Plot)
A histogram is used to show the shape and distribution of the numerical variable. For categorical variables, it shows the count of the categories present in the variable.
From both histograms, it is shown that the HP variable is quite distributed. It is a little bit tilted in right. That means it is slightly right-skewed but normally distributed. However, the price variable is highly skewed.
Histogram for Categorical Variable
This is the countplot for Make Variable. Every bar shows the count of the category present in the dataset.
Outliers are the values which are significantly different from other values/observations. An outlier can create major issues in modelling. So it is necessary to find outliers and treat them.
Outliers can be detected by using boxplot. Boxplot depicts the variable distribution using quartile. It is also known as a box and whiskers plot.
All the above boxplots show that there are many outliers present in price and c_mpg variables. In Cylinders variable only 4 observations are outliers.
According to the box plot, any observation which is out of the range of Q1 (25 percentile) and Q3 (75 percentile) or IQR (Inter quartile range), is observed as an outlier.
If there are lots of outliers present in the dataset, then the treatment of outliers is necessary. There are methods like flooring, capping which can be used to impute outliers.
Correlation is calculated to find out the intensity of the relationship between 2 variables. Correlation ranges from -1 to 1. -1 correlation value suggests the strong negative relationship and 1 shows a strong positive relationship. 0 means there is no relation between 2 variables.
From the above correlation plot, it can be inferred that there are many variables which are strongly related to each other. For Example, the correlation value between c_mpg and h_mpg is 0.85 which is near to 1. That means there is a strong positive relationship between them. Likewise, Cylinders and c_mpg have a negative relationship.
Pairplot is used to find out the relationship between variables. It plots the scatter plot between each variable. Scatter plot can also be used independently. But pairplot will give the relationship plot among all the numerical variables in one line.
All the above steps are part of EDA. This is not the end of EDA. All the steps above performed are the basics which should be performed to analyse the data before doing feature engineering or modelling.
EDA is one of the important steps during the whole process of data science. It is said that most of the time of model building goes into EDA and feature engineering. If you want to create a big set up of information from the data, you need to do an extensive EDA.4