Subject: DSCI521-Analysis-Interpretation-project

Team members:

1. Name: Madhu Bandru
2. Name: Sudharshan Rao B
3. Name: Spandana Bendi
4. Name : Divya Lakshmi Narayana

INTRODUCTION

• We have taken a dataset from Kaggle.com which has used car informations
• We used this data to predict the price based on relevant features available in the dataset
• The database has been created with data on their sales upto previous year
• We are using machine learning techniques for price prediction
• This dataset is going to provide the estimation quote of price based on data features entered by the user
• It will help to guide the users to estimate the price for the used cars

SYSTEM REQUIREMENTS AND PRE-REQUISITES

• OS: Mac, Linux or Windows
• Software: Python 3.0 or later
• OS :Mac,Linux or Windows
• Python3.0 or above
• Jupyter Lab
• Django

ABOUT THE DATASET

Fields:

• dateCrawled- series object
• name-series object
• seller-series object
• offerType-series object
• price-series int
• abtest-series object
• vehicleType - series object
• yearOfRegistration- series int
• gearbox- series object
• powerPS- series int
• model- series object
• kilometer- series int
• monthOfRegistration- series int
• fuelType - series object
• brand- series object
• notRepairedDamage - series object
• dateCreated - series object
• nrOfPictures- series int
• postalCode- series int
• lastSeen - series object

Models implemented:

• Linear Regression
• K-Nearest Neighbors
• MLP Regressor
• Decision tree
• Random Forest

Attributes used:

• Price
• vehicleType
• gearbox
• powerPS
• model
• kilometer
• fuelType
• brand
• postalCode
• vehicleAge

DATA CLEANSING

• Reformatted existing columns for easier analysis.
• NaN, space values filled with mean, median and mode.
• Conversion of few columns to date-time and integer values as needed for analysis.

STEPS TO EXECUTE THE CODE

• pip install below packages:
  • Pandas
  • Numpy
  • Sklearn
  • matplotlib
  • seaborn
• Simply place the data in a separate directory and name it as 'Data'
• run the .ipynb file and execute the code

UI

• We have designed the UI based on the above model in a clickable format and used it to predict the price

Conclusion

• We evaluated used-car price prediction using Kaggle dataset which gives an accuracy of 93% with RandomForest Model.
• The most relevant features used for this prediction are price, year, Power_ps, and vehicleAge by filtering out outliers and irrelevant features of the dataset.
• This model gives us good accuracy when compared to the other models.

FUTURE WORK

• To collect data for more cars
• More information of cars in the current dataset
• To adjust model hyper-parameters, by using RandomizedSearchCV