Assignment 3: Final report
‘Used Volkswagen cars -Belarus‘
,Table of contents
One-page executive summary................................................................................................................. 3
Introduction............................................................................................................................................. 4
1. Data description .................................................................................................................................. 5
1.1 Data source & reliability ................................................................................................................ 5
1.2 Composition of the dataset ........................................................................................................... 5
1.3 Data quality ................................................................................................................................... 6
2. Correlation analysis ............................................................................................................................. 7
2.1 Simple Correlations ....................................................................................................................... 7
2.2 Pair-wise correlations .................................................................................................................... 8
3. Predictive modeling ............................................................................................................................. 9
3.1 KPI Drivers ..................................................................................................................................... 9
3.2 Prediction model performance ................................................................................................... 10
3.3 What-Ifs ....................................................................................................................................... 11
Conclusion ............................................................................................................................................. 12
Bibliography........................................................................................................................................... 13
,One-page executive summary
Working as a junior data analyst for the company 'Timmermans NV', I was commissioned to
determine the attractiveness of the used Volkswagen car market in Belarus.
The goal is to investigate which factors have a major influence on the price of the cars, this is the KPI.
A dataset of 4 244 used Volkswagen cars with 10 factors per car has been analyzed with advanced
software from DataStories.com. The general data health is 81.8%. Correlation analysis and predictive
modeling has been performed.
Correlation analysis:
Insights
• Insight 1: The factor ‘year_produced’ has 85% mutual information with the KPI. The newer
the car, the more expensive it will be. This is a very strong positive linear correlation.
• Insight 2: The factor ‘odometer_value’ has 39.0% mutual information content with the price
of the cars. The higher the kilometers on the odometer the less expensive the car is. This is a
negative linear correlation.
• Insight 3: The transmission system ( automatic or manual ) has 34% mutual information
content with the price of the cars. The automatic cars are way more expensive in general.
Predictive modeling:
To build a predictive model with a prediction accuracy of 91.0% Datastories uses 3 KPI drivers
1. year_produced importance of 89.9%
2. engine_capacity importance of 7%
3. body_type importance of 3.1%
Insights
• Insight 1: The value of cars decreases very fast in the first 10 years, after 10 years the price
decreases less rapidly.
• Insight 2: Between 1.5- and 2.75-liters engine capacity, the value of the cars increases very
quickly as the capacity increases. If the engine capacity is higher than 2.75 liters this will not
give more value to the car, it may even decrease.
• Insight 3: Minibuses, vans, pickups are more expensive in general. Hatchbacks, SUV’s,
liftbacks are cheaper in general.
In general, the prediction model is very accurate for cheaper cars, but the accuracy decreases, and
the outliers increase as the cars become more expensive.
With the clear insights and the accurate prediction model, Timmermans NV has a clear view on the
used Volkswagen car market in Belarus. Thanks to the predictive model the company can estimate
accurately which Volkswagen cars are under -and overvalued. This is perfect for resellers. The profit
margin per car will be higher what makes the business more lucrative.
,Introduction
The company: ‘Timmermans NV’ I work for as a junior data analyst, is currently selling used
Volkswagen cars in Belgium. They asked me to determine the attractiveness of the Volkswagen car
market in Belarus. As a car dealer it is very important to accurately estimate the value of used cars.
This way you will increase the profit per car and make your business more lucrative. (Buy low, sell
high.)
To solve the business problem, I will use the data from 4 244 used Volkswagen cars. The data was
web-scraped in Belarus on the 2nd of December 2019. The dataset exists of 10 different factors per
car. The Key Performance indicator used is ‘Price_usd’, it reflects the value of the used cars.
Firstly, I will perform correlation analysis to gain interesting insights in the Volkswagen used car
market in Belarus. This is how the attractiveness of the market is determined and analyzed.
Afterwards I apply predictive modeling to the dataset using Datastories.com so the Volkswagen
dealer can determine which cars are under- and overvalued.
If the Volkswagen used car market in Belarus proves interesting, the company will expand to the
Belarusian market. With the help of the predictive modeling it is possible to determine what are
good prices to buy and sell cars. The model allows the business to be more lucrative.
, 1. Data description
1.1 Data source & reliability
To solve the business problem, I will use a dataset found on Kaggle (Lepchenkov, 2019), one of the
largest online communities for data scientists and machine learning. The data has been collected
from the most well-known websites, selling used cars in Belarus. The data from 38 531 cars were
web-scraped in Belarus on the 2nd of December 2019 . The dataset exists of 11 different factors per
car.
I adjusted the database before uploading it to Datastories. Only the Volkswagen cars are still in the
dataset, the rest has been deleted. The data consists of 4 244 Volkswagen cars which are listed
online for sale. The set is very recent and relevant, it gives a clear reflection of the used car market in
Belarus.
1.2 Composition of the dataset
The KPI of interest is ‘price_usd’, the influence of the following factors on the KPI is analyzed:
1. Model_name TXT
2. Transmission ( Automatic / Manual ) BIN
3. Color ( Black, white, silver, … ) CAT
4. Odomoter_value ( 0km – 900 000km) NUM
5. Year_produced ( 1975 – 2019 ) NUM
6. Engine_fuel ( Gasoline, diesel, gas, hybrid ) CAT
7. Engine_ type ( Gasoline / Diesel ) BIN
8. Engine_capacity ( 1,00 liter – 5.00 liter ) NUM
9. Body_type ( Sedan, SUV, Coupe, minibus, … ) CAT
10. Drivetrain ( Front / Rear / All ) CAT
On the graph above, we see the distribution of the cars according to the KPI, price in USD (x-axis).
Most cars (1 804 from 4 244 cars) cost between 0 and 4400.8 USD. In general, quite a lot of cars cost
up to 13 200 USD. (90.95%)
‘Used Volkswagen cars -Belarus‘
,Table of contents
One-page executive summary................................................................................................................. 3
Introduction............................................................................................................................................. 4
1. Data description .................................................................................................................................. 5
1.1 Data source & reliability ................................................................................................................ 5
1.2 Composition of the dataset ........................................................................................................... 5
1.3 Data quality ................................................................................................................................... 6
2. Correlation analysis ............................................................................................................................. 7
2.1 Simple Correlations ....................................................................................................................... 7
2.2 Pair-wise correlations .................................................................................................................... 8
3. Predictive modeling ............................................................................................................................. 9
3.1 KPI Drivers ..................................................................................................................................... 9
3.2 Prediction model performance ................................................................................................... 10
3.3 What-Ifs ....................................................................................................................................... 11
Conclusion ............................................................................................................................................. 12
Bibliography........................................................................................................................................... 13
,One-page executive summary
Working as a junior data analyst for the company 'Timmermans NV', I was commissioned to
determine the attractiveness of the used Volkswagen car market in Belarus.
The goal is to investigate which factors have a major influence on the price of the cars, this is the KPI.
A dataset of 4 244 used Volkswagen cars with 10 factors per car has been analyzed with advanced
software from DataStories.com. The general data health is 81.8%. Correlation analysis and predictive
modeling has been performed.
Correlation analysis:
Insights
• Insight 1: The factor ‘year_produced’ has 85% mutual information with the KPI. The newer
the car, the more expensive it will be. This is a very strong positive linear correlation.
• Insight 2: The factor ‘odometer_value’ has 39.0% mutual information content with the price
of the cars. The higher the kilometers on the odometer the less expensive the car is. This is a
negative linear correlation.
• Insight 3: The transmission system ( automatic or manual ) has 34% mutual information
content with the price of the cars. The automatic cars are way more expensive in general.
Predictive modeling:
To build a predictive model with a prediction accuracy of 91.0% Datastories uses 3 KPI drivers
1. year_produced importance of 89.9%
2. engine_capacity importance of 7%
3. body_type importance of 3.1%
Insights
• Insight 1: The value of cars decreases very fast in the first 10 years, after 10 years the price
decreases less rapidly.
• Insight 2: Between 1.5- and 2.75-liters engine capacity, the value of the cars increases very
quickly as the capacity increases. If the engine capacity is higher than 2.75 liters this will not
give more value to the car, it may even decrease.
• Insight 3: Minibuses, vans, pickups are more expensive in general. Hatchbacks, SUV’s,
liftbacks are cheaper in general.
In general, the prediction model is very accurate for cheaper cars, but the accuracy decreases, and
the outliers increase as the cars become more expensive.
With the clear insights and the accurate prediction model, Timmermans NV has a clear view on the
used Volkswagen car market in Belarus. Thanks to the predictive model the company can estimate
accurately which Volkswagen cars are under -and overvalued. This is perfect for resellers. The profit
margin per car will be higher what makes the business more lucrative.
,Introduction
The company: ‘Timmermans NV’ I work for as a junior data analyst, is currently selling used
Volkswagen cars in Belgium. They asked me to determine the attractiveness of the Volkswagen car
market in Belarus. As a car dealer it is very important to accurately estimate the value of used cars.
This way you will increase the profit per car and make your business more lucrative. (Buy low, sell
high.)
To solve the business problem, I will use the data from 4 244 used Volkswagen cars. The data was
web-scraped in Belarus on the 2nd of December 2019. The dataset exists of 10 different factors per
car. The Key Performance indicator used is ‘Price_usd’, it reflects the value of the used cars.
Firstly, I will perform correlation analysis to gain interesting insights in the Volkswagen used car
market in Belarus. This is how the attractiveness of the market is determined and analyzed.
Afterwards I apply predictive modeling to the dataset using Datastories.com so the Volkswagen
dealer can determine which cars are under- and overvalued.
If the Volkswagen used car market in Belarus proves interesting, the company will expand to the
Belarusian market. With the help of the predictive modeling it is possible to determine what are
good prices to buy and sell cars. The model allows the business to be more lucrative.
, 1. Data description
1.1 Data source & reliability
To solve the business problem, I will use a dataset found on Kaggle (Lepchenkov, 2019), one of the
largest online communities for data scientists and machine learning. The data has been collected
from the most well-known websites, selling used cars in Belarus. The data from 38 531 cars were
web-scraped in Belarus on the 2nd of December 2019 . The dataset exists of 11 different factors per
car.
I adjusted the database before uploading it to Datastories. Only the Volkswagen cars are still in the
dataset, the rest has been deleted. The data consists of 4 244 Volkswagen cars which are listed
online for sale. The set is very recent and relevant, it gives a clear reflection of the used car market in
Belarus.
1.2 Composition of the dataset
The KPI of interest is ‘price_usd’, the influence of the following factors on the KPI is analyzed:
1. Model_name TXT
2. Transmission ( Automatic / Manual ) BIN
3. Color ( Black, white, silver, … ) CAT
4. Odomoter_value ( 0km – 900 000km) NUM
5. Year_produced ( 1975 – 2019 ) NUM
6. Engine_fuel ( Gasoline, diesel, gas, hybrid ) CAT
7. Engine_ type ( Gasoline / Diesel ) BIN
8. Engine_capacity ( 1,00 liter – 5.00 liter ) NUM
9. Body_type ( Sedan, SUV, Coupe, minibus, … ) CAT
10. Drivetrain ( Front / Rear / All ) CAT
On the graph above, we see the distribution of the cars according to the KPI, price in USD (x-axis).
Most cars (1 804 from 4 244 cars) cost between 0 and 4400.8 USD. In general, quite a lot of cars cost
up to 13 200 USD. (90.95%)
