https://github.com/aabbtree77/uci-marketing-analysis-cart
UCI bank marketing data analysis with decision trees (CART).
https://github.com/aabbtree77/uci-marketing-analysis-cart
cart chatgpt commerce conversion-rate data-analysis decision-trees deepseek grok kovnatsky marketing-analytics miniconda scikit-learn-python uci-machine-learning
Last synced: 24 days ago
JSON representation
UCI bank marketing data analysis with decision trees (CART).
- Host: GitHub
- URL: https://github.com/aabbtree77/uci-marketing-analysis-cart
- Owner: aabbtree77
- License: mit
- Created: 2025-04-20T17:21:07.000Z (about 1 month ago)
- Default Branch: main
- Last Pushed: 2025-04-20T17:42:35.000Z (about 1 month ago)
- Last Synced: 2025-04-20T18:35:04.804Z (about 1 month ago)
- Topics: cart, chatgpt, commerce, conversion-rate, data-analysis, decision-trees, deepseek, grok, kovnatsky, marketing-analytics, miniconda, scikit-learn-python, uci-machine-learning
- Language: Python
- Homepage:
- Size: 74.2 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
Toretsk (Ukraine, April 2025)
[...](https://www.reddit.com/r/UkraineRussiaReport/comments/1k2d20a/ua_pov_birdseye_view_of_what_remains_of_toretsk/)
## The UCI Bank Marketing Campaign Decision Tree Analysis
This project analyzes [the UCI Bank Marketing Dataset](https://archive.ics.uci.edu/dataset/222/bank+marketing) using CART to predict customer subscription (a binary variable). A conversion rate is the average of the subscription value for a chosen data subset (market segment).
## Python3 and miniconda (Ubuntu 22.04)
```bash
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh
```
The last step of miniconda (155MB) install: "Do you wish to update your shell profile to automatically initialize conda?" I have chosen "No", and simply initialize it manually:
```bash
source /home/tokyo/miniconda3/etc/profile.d/conda.sh
```
## Dependencies
Environment:
```bash
conda create -n banktree
conda info --envs
# conda environments:
#
base /home/tokyo/miniconda3
banktree /home/tokyo/miniconda3/envs/banktree
conda activate banktree
```
Dependencies:
```bash
conda install python=3.13
conda install pandas scikit-learn
conda install -c conda-forge ucimlrepo certifi
conda install requests tabulate
```
Exit and removal:
```bash
conda deactivate
conda env remove --name banktree
rm -rf /home/tokyo/miniconda3/envs/banktree
conda clean --all
```
## Grok
"Give me the script which loads the UCI Bank Marketing Dataset, splits 20% into testing, builds CART, outputs training set sample number, accuracy, conversion rate, same for testing. Also, output top ten groups based on job and education with highest conversion rates and show sample numbers, nothing else."
```bash
python main_grok.py
Training set:
Sample number: 36168
Accuracy: 1.0000
Conversion rate: 0.1161
Testing set:
Sample number: 9043
Accuracy: 0.8734
Conversion rate: 0.1206
Top 10 groups based on job and education with highest conversion rates:
Job Education Conversion Rate Sample Number
student primary 0.363636 44
student secondary 0.297244 508
retired tertiary 0.275956 366
student tertiary 0.264574 223
retired primary 0.223899 795
retired secondary 0.210366 984
unemployed tertiary 0.193772 289
admin. tertiary 0.173077 572
blue-collar tertiary 0.161074 149
self-employed tertiary 0.160864 833
```
## deepseek
"Give me the script which loads the UCI Bank Marketing Dataset, splits 20% into testing, builds CART, outputs training set sample number, accuracy, conversion rate, same for testing. Also, output top ten groups based on job and education with highest conversion rates and show sample numbers, nothing else."
"The link is https://archive.ics.uci.edu/static/public/222/bank+marketing.zip, and it's a zip file, not csv! Inside bank+marketing.zip there are bank.zip and bank-additional.zip. Inside bank.zip there is bank.csv (around 460 KB), bank-full.csv (around 4.6MB) and bank-names.txt 3.9 KB. Inside bank-additional.zip there is bank-additional folder inside it bank-additional.csv (around 584KB), bank-additional-full.csv (~5.8MB), and bank-additional-names.txt (~5.5KB)."
```bash
python main_deepseek.py
Training Set
Samples: 32,950
Accuracy: 100.00%
Conversion Rate: 11.24%
Testing Set
Samples: 8,238
Accuracy: 88.69%
Conversion Rate: 11.35%
Conversion Rate Ranges:
Max: 35.35%
Min: 18.64%
Top 10 Job/Education Groups:
| | Conversion_Rate | Samples |
|:-------------------------------------|------------------:|----------:|
| ('student', 'basic.9y') | 0.353535 | 99 |
| ('student', 'unknown') | 0.353293 | 167 |
| ('retired', 'unknown') | 0.336735 | 98 |
| ('student', 'high.school') | 0.319328 | 357 |
| ('retired', 'basic.4y') | 0.309883 | 597 |
| ('retired', 'professional.course') | 0.236515 | 241 |
| ('retired', 'university.degree') | 0.231579 | 285 |
| ('retired', 'high.school') | 0.224638 | 276 |
| ('student', 'university.degree') | 0.205882 | 170 |
| ('housemaid', 'professional.course') | 0.186441 | 59 |
```
## Notes
* As Leo Breiman has noted himself in 2001, CART is not the most accurate method.
* CART is great in that it handles any data (missing, mixing continuous with nominal), and is automatic. It is also fast: no inverses, no learning, no GPUs needed. Ideal for rough estimates.
* I would not spend too much time on the generated trees, clusters/rules, variable importance.
* pip is horrible, but conda solves the problem. Jupyter Notebook is not that useful.
* ChatGPT, deepseek, and Grok are great for such scripts, but one needs to debug/iterate.
* [Artiom Kovnatsky](https://www.artiomkovnatsky.com/) uses CART in real-world commercial projects.
## References
[A Conversation with Leo Breiman (2001)](https://projecteuclid.org/journals/statistical-science/volume-16/issue-2/A-Conversaton-with-Leo-Breiman/10.1214/ss/1009213290.full)