Problem Statement

Aims

1. Convert the multi-class problems into binary classification tasks.
2. Predict whether a person is a consumer of chocolate and magic mushroom.
3. Choose the best and worst classifiers for each dataset.
4. Explain AI models in a scientific manner which should be convincable to non-technical people.
5. Implement models with Semi-Supervised Learning.

Preview

Comparing a Pipeline of 6 classifiers on 2 datasets Explainable AI Semi-Supervised Learning

Dataset: Drug Consumption Analysis Dataset

The dataset can be found at this link: https://archive.ics.uci.edu/dataset/373/drug+consumption+quantified.

Description of the Dataset

• Contains a row identifier, 12 features describing the user data, and 18 classification problems related to using 18 different drugs.
• For each drug, it indicates whether a person has 'never used', 'used over a decade ago', 'used in the last decade', 'used in the last year', 'used in the last month', 'used in the last week', or 'used in the last day'.

Implementation Details

• Split the Dataset and Perform Feature Engineering.
• Perform Supervised Learning using a pipeline of 6 classifiers.
• Identify potential issues in the dataset / the classifier itself.
• Provide results from Evaluation with some useful plots and metrics.
• Summarize the analysis in a report.
• Explain whether certain classifiers make trustable predictions, with the calculation of SHAP values and some visualization plots.
• Prepared labelled and unlabelled data, Implemented and Compared different semi-supervised learning algorithms based on the gradient boosting classifier from assignment 1.

Project Structure

• You should expect some reports in .pdf format at the root level.
• The report for the project is inside the project folder, with name: "Project - Semi Supervised Learning/KeycodeExplaination.pdf"
• Please expand the folder at the root level to view codes.
• This project branches out the analysis into 9 notebooks:
  1. Modelling - please check the file CSI5155 Assignment 1 Modelling Part- Kelvin Mock 300453668.ipynb
  2. Evaluation - please check the file CSI5155 Assignment 1 Evaluation Part - Kelvin Mock 300453668.ipynb
  3. Calculation of SHAP Values - please check the file CSI5155 Assignment 2 - Kelvin Mock 300453668.ipynb
  4. Visualizing the SHAP Values - please check the file CSI5155 Assignment 2 Plots - Kelvin Mock 300453668.ipynb
  5. Baseline Model (Gradient Boosting classifier) - CSI5155 Project - baseline.ipynb
  6. Self Training method applied on baseline model - CSI5155 Project - Self Training.ipynb
  7. Co-Training method applied on baseline model - CSI5155 Project - Co Training.ipynb
  8. Semi-Boost method applied on baseline model - CSI5155 Project - Semi Boost.ipynb
  9. Label Spreading method applied on baseline model - CSI5155 Project - Label Spreading.ipynb
• Models are data dumped into several .pkl files from time-to-time in different phases to maintain the code's maintainability.
• The training sets and test sets are also data dumped into several .pkl files.
• choc directory shows data dumped files related to the Chocolate dataset (which is split from the original dataset).
• mushrooms directory shows data dumped files related to the Mushrooms dataset (which is also split from the original dataset).