Car Evaluation Analysis

Authors: Danish Karlin Isa, Nicholas Varabioff, Ximin Xu, Zuer Zhong

This project is part of the coursework for DSCI 522 Data Science Workflows, a course of the Master of Data Science program at the University of British Columbia.

About

This project attempts to predict the level of acceptability of cars using machine learning methods. Using a 1997 dataset, the influence of various attributes of a new car on its acceptability among customers is analysed. These attributes include:

• The buying price of the car
• Maintenance costs
• Number of doors
• Passenger capacity
• Boot size
• Safety ratings

This analysis aims to identify the key factors that determine whether a car is considered acceptable, good, or unacceptable according to standardized criteria. To achieve this, several common machine learning models were explored. The SVM RBF classifier was identified as the best-performing model, achieving a test accuracy of 0.952. On the 346 test data cases, it correctly predicted the targets of 343 examples, with only 3 misclassifications. The SVM RBF model demonstrated exceptional performance, as shown by its confusion matrix, classification reports, and high precision, recall, and F1 scores. However, a slight decrease in classification precision was observed for the “good” category, alongside a relatively lower recall score of 0.86, indicating occasional classification errors.Nonetheless, the results obtained from this analysis further exemplifies the ability of the SVM RBF model in handling nonlinear decision boundaries. This makes the SVM RBF model a solid choice for this project.

The dataset used in this project is the Car Evaluation Database created by M. Bohanec and V. Rajkovic in the early 1990s. It was sourced from the UCI Machine Learning Repository and is publicly available for research and can be found in the UCI Machine Learning Repository.

Report

The final report can be found here.

Usage

To run this project, follow these steps from the root of this repository:

1. If you are running this project for the first time, run the following from a Command-Line Interface:

conda-lock install --name Car_Evaluation_Analysis conda-lock_<your_operating_system>.yml

2. To run the analysis, run the following command:

jupyter lab

Open notebooks/Car_Evaluation_Analysis.ipynb in Jupyter Lab and under Switch/Select Kernel choose "Python [conda env:car_evaluation_analysis]".

Next, under the "Kernel" menu click "Restart Kernel and Run All Cells...".

Dependencies

• conda
• conda-lock
• jupyterlab
• nb_conda_kernels
• Python and packages listed in environment.yml

License

This dataset is licensed under a Creative Commons Attribution 4.0 International (CC BY 4.0) license. This allows for the sharing and adaptation of the datasets for any purpose, provided that the appropriate credit is given.The software code contained within this repository is licensed under the MIT license. See the license filefor more information.

References

• Bohanec, M. (1988). Car Evaluation [Dataset]. UCI Machine Learning Repository. (https://doi.org/10.24432/C5JP48).
• Makki, S., Mustapha, A., Kassim, J. M., Gharayebeh, E. H., & Alhazmi, M. (2011, April). Employing neural network and naive Bayesian classifier in mining data for car evaluation. In Proc. ICGST AIML-11 Conference (pp. 113-119).
• Potdar, K., Pardawala, T. S., & Pai, C. D. (2017). A comparative study of categorical variable encoding techniques for neural network classifiers. International journal of computer applications, 175(4), 7-9.
• Tanveer, M., Gautam, C., & Suganthan, P. N. (2019). Comprehensive evaluation of twin SVM based classifiers on UCI datasets. Applied Soft Computing, 83, 105617.