Reference no: EM133179254

MLN 601 Machine Learning - Torrens University Australia

Assessment - Classification

Learning Outcome 1: Apply learning algorithms to perform machine learning tasks.

Learning Outcome 2: Implement practical machine learning: data pre-processing, analysis, model selection, and interpret the results.

Learning Outcome 3: Communicate clearly and effectively using the technical language of machine learning to a range of stakeholders

Task Summary

Assessment 1 considered a wine data set as a regression task. This brief revisits the data set as a classification task. In this Assessment, you will use a decision tree Machine Learning (ML) algorithm to analyse data and draw conclusions. To help you create and document this ML model and the results, you will follow the end-to-end CRoss-Industry Standard Process for Data Mining (CRISP-DM) (Chapman et al., 2000) methodology. Further, to guide you through the analysis, the development of your report and model, and the writing of your report and 7-10 minute presentation, a template for your Jupyter Notebook has been provided with comments. Your presentation should touch on the key steps of the template, including the lessons you learned and your experiences.

Please refer to the Task Instructions (below) for further details on how to complete this task.

Context
In addition to giving you an opportunity to complete a ML exercise, this Assessment also gives you an opportunity to practice hyperparameter tuning using the really useful scikit-learn library. In your future workplaces, you will often be expected to perform similar exercises using suitable data sets with different machine learners and tune the hyperparameters. Model building requires you to revise parameters and tune them for the next model run.

For this Assessment, the practice data set is available from the UCI ML repository, which contains nearly 500 real-world data sets. Your focus will be on the wine quality data set. This data set provides wine quality data across 11 traits, including acidity, residual sugar and alcohol concentration. Importantly, this Assessment requires you to develop a model to predict wine quality on a score between 1 to 10.
You will revisit this data set to complete a classification task. To achieve this, you will setup a categorical variable with two categories. Thus, you will be required to allocate levels for your wine quality (the dependent variable) to assign either a ‘low' quality (1) (below the value of 6) or a ‘high' quality (0) (below the selected value of 6). You will use this binary classification to help generate a prediction model for high or low quality wine using decision tree algorithms.

Follow the steps of the CRISP-DM model using the template CRISP_DM _Template_(assessment_2_ classification).ipynb to document and develop your ML model. At the modelling stage, you should practice tuning the hyperparameters for the decision tree to ascertain the effects on the model and determine the optimal performance using the AUC-ROC curve.

Task Instructions

You will use your Jupyter Notebook on the Microsoft Azure ML platform or Google Colab and Python
3.6 as the language for all three assessments.
Ultimately, the Notebook will contain both your ML code, data and report documentation.
Your Assessment will be evaluated based on the major stages of the CRISP-DM process as set out in the Notebook template with prompts. The process comprises:
1. Business Understanding;
2. Data Understanding;
3. Data Preparation;
4. Modelling;
5. Evaluation; and
6. Deployment.
The six multi-step stages of the CRISP-DM must be undertaken to complete this Assessment. Note: For ease of working and to complete this Assessment, you should document what you are doing in your Notebook as you progress through the activities (e.g., the steps undertaken and the rationale for the selection of the code). The template will prompt you on how to work through the end-to-end ML process.

Stage 1: Business Understanding
1. This section serves as an introduction. You should write a clear and concise narrative expressing what you are trying to achieve. Think in terms of ML; for example, the prediction algorithm, the data set selected, what you are seeking from the data set and how you intend to understand the value of your prediction capability.
2. Assess the current situation. See 1.1 of the CRISP-DM template (1.1).

Stage 2: Data Understanding
1. Acquire the relevant wine quality data set from the UCI repository for your prediction model. Explicitly specify the data source by providing a specific link and the name of the data set (e.g., red wine, white wine or both) and the method of acquisition (e.g., direct from the URL or a download of the .csv file). The steps taken need to be clearly stated. (2.1).
2. Read this data set into your Notebook. (2.1).
3. Describe the data set inclusive of variables, units and levels. (2.2).
4. Verify the data quality by analysing the data set for structure and missing data. (2.3).
5. Conduct an initial data exploration using data visualisation, reporting and querying of the data. (2.4).
6. Use the pairplot function in seaborn to determine the relationship, if any, between the variables. Include the output or the visualisation of the pairplot function in your Notebook and comment on it. (2.4.2).

Stage 3: Data Preparation
1. Select the data that you will use for the analysis. (3.1).
2. Clean the data you have selected to improve the quality of the data. (3.2).

Stage 4: Modelling
1. For this Assessment, you are only required to consider one classification modelling technique (e.g., a decision tree).
2. Import the decision tree model in your code. (4.1).
3. Record any modelling assumptions. (4.2).
4. Run your model over the data set. (4.3).
5. Record the parameter settings, your rationale for your choice of values and the actual model generated. (4.3).
6. Revise any parameter settings for subsequent model runs. Document all the revisions until the best model is reached. (4.4).
7. Assess the model or models according to the performance measurement set to meet your evaluation criteria. The AUC-ROC curve is useful for the performance measurement of classification.
8. Revise any parameter settings for subsequent model runs. Document all the revisions until the best model is reached. (4.4).

Stage 5: Evaluation
1. Assess the ML results. Ensure you include a statement as to whether the model meets the initial objective.

Stage 6: Deployment
1. For this Assessment, you are not required to deploy your model. For this stage, simply include any lessons that you learned and that you wish to share in relation to the things that went right and wrong, the areas in which you did well and in which you could improve. You can also detail any of your other experiences in completing this Assessment.

Stage 7: Presentation
1. Once complete, you should setup the Jupyter Notebook for screen recording. You are required to make a screen recording of your Jupyter Notebook and a webcam video of yourself narrating for 7-10 minutes. You should specify your name and any other student details at the beginning. Work your way through the Notebook as you discuss the key aspects of the CRISP-DM steps, the lessons you learned and any other experiences.

2. A wide variety of tools are available to record videos of a webcam and screen simultaneously (i.e., picture in picture). In this case, your video will show you discussing your Notebook on your screen. Use the large screen for your Notebook. Available tools include the inbuilt recorder for Windows 10, Quicktime on Apple, fluvid.com, panopto.com or Zoom. Owing to the size of the video file, you will be submitting the URL for the file. Practice the presentation beforehand to ensure clarity and conciseness.