https://github.com/alexduvalinho/explainable-artificial-intelligence-xai-
Master Research Project, University of Warwick
https://github.com/alexduvalinho/explainable-artificial-intelligence-xai-
Last synced: about 2 months ago
JSON representation
Master Research Project, University of Warwick
- Host: GitHub
- URL: https://github.com/alexduvalinho/explainable-artificial-intelligence-xai-
- Owner: AlexDuvalinho
- Created: 2018-09-23T16:03:24.000Z (about 7 years ago)
- Default Branch: master
- Last Pushed: 2020-05-13T12:51:52.000Z (over 5 years ago)
- Last Synced: 2025-04-22T08:43:29.816Z (6 months ago)
- Language: Jupyter Notebook
- Homepage:
- Size: 21.9 MB
- Stars: 4
- Watchers: 0
- Forks: 3
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Explainable Artificial Intelligence (XAI)
**Master Research Project, University of Warwick**
The **paper** attached above ('Research Project.pdf') examines Explainable Artificial Intelligence (XAI) and sheds light on making machine learning models more interpretable. The current inability of humans to explain these models creates a barrier to the adoption of machine learning that becomes extremely problematic given the range of opportunities it creates for the society. To tackle this issue, this paper positions itself as a scientific guide for interpretability, filling in a major gap in the literature. More precisely, after setting the foundations for this concept, it provides a unified mathematical view of the current main explanation methods, which is vital to truly comprehend both the purpose and functioning of the technique utilised. This rigorous theoretical approach is then completed with an application on a customer churn use case for a telecom company. Its analysis illustrates how to implement, interpret and ideally combine these methods to obtain the best possible understanding of a real-world problem. Finally, to achieve an even more complete comprehension, this paper emphasises the need to search for some alternative approaches to the problem. In this case, survival analysis is chosen, applied and its theory expanded.
This **repository** supports the above paper by tackling the application mentioned, meaning the customer churn use-case. It includes a deeply commented version of the code used to visualise the data and to implement the model, the interpretation methods and survival analysis. All results are either easily deducible from the IPython notebook files' output or specified in the comments. The main conclusions drawn from these results are then summarised in the paper (sections 3.3, 4.7, 5.6 respectively for model construction, interpretability and survival analysis).
The **dataset** considered looks at fictitious customers who churned during the last month in the telecommunication company named Telco. The dataset was taken on IBM's website; it contains 7043 observations and 21 variables. Each row represents a unique client and each column contains information about his account (contract, payment method, etc.), his demography (gender, age, etc.) and his services (phone, multiple lines, internet, etc.). The exhaustive list of variables used in this analysis is included in Appendix C.