辅导 MGRC20007 business analyse (Machine Learning for Data-driven Business Decision-making 2025) Asses

MGRC20007 business analyse (Machine Learning for Data-driven Business Decision-making 2025)

Assessment 2 - Individual Report

Assessment 2 -- Individual Report (70%)

All the key details in one place. This Brief is available in other formats via the I Download Alternative FormatsI button above or on request to your tutor or School Oﬃce.

Deadline	4th December 2025 13:00 GMT.
Length	2,000 words
Assessment type	Individual report
Submission format	Turnitin

Assessment task outline

Individual coursework - Task outline

1. Revisit the group churn solution and restate its decision context, baseline, and limits.

2. Set a clear improvement goal for your individual work with success metrics and constraints.

3. Audit the data you will use, document any changes from the group version, and justify them.

4. Design improvements to the model pipeline. Consider feature work, tuning, and at least one alternate algorithm.

5. Validate rigorously with cross validation and a final holdout. Compare against the group baseline.

6. Explain model behaviour and drivers. Use appropriate interpretability methods and connect to business logic.

7. Check fairness across relevant groups, report disparities, and propose mitigations.

8. Apply your refined model to a concrete bank scenario and quantify expected impact and risk.

9. Reflect critically on what you learned, trade offs made, and what you would change next time.

10.Present a professional report with labelled figures, sound references, and a concise appendix that enables reproducibility.

Assessment-specific requirements

• Report length is 2,000 words +/- 10%, excluding references and appendices.

• Focus on application to a business decision rather than theory in isolation.

• Start from your group churn model, state the baseline, and show clear individual improvements.

• Use the same dataset or a justified variant, document any changes, and avoid unapproved sensitive data.

• Reproducibility is required. Submit runnable code or notebooks, a short readme with run steps, fixed random seed, and package versions.

• Evaluate with cross validation and a final holdout. Report accuracy, precision, recall, F1, ROC AUC, and at least one business metric such as expected retention revenue or uplift at top decile.

• Provide interpretability using feature importance or SHAP and include at least two manager friendly visuals.

• Conduct a fairness check across at least two relevant customer groups, report disparities, and propose mitigations.

• Apply the refined model to one concrete bank scenario, quantify expected impact and risk, and outline an A B test and monitoring plan.

• Use Harvard referencing throughout with 10-20 sources. Do not cite lecture slides. Label all tables and figures with captions, for example Figure 1, and refer to them in the text.

• Follow the unit Generative AI policy. Include an AI use statement in the appendix naming tools, purposes, prompt summaries, and verification steps. Do not submit undisclosed AI generated text, code, or fabricated results. You may be invited to a short viva to confirm understanding.

• Appendices must include a pipeline diagram, model settings, feature list, brief data dictionary, fairness summary, AI use statement, and code listing or link to a private repository with environment file.

• File format and naming. Submit a single PDF for the report. Name the file with student ID and unit code only.

Generative AI rules

• You may use generative AI for idea generation, outlining, editing for clarity, small code snippets, debugging, docstrings, visual drafts, and slide polish. You must remain the author.

• Do not use generative AI to write whole sections of the report or slides, to run the analysis in place of you, to fabricate data, figures, metrics, or references, or to generate undisclosed code.

• Include an AI use statement in the appendix. Name the tools used, what they were used for, short summaries of prompts or queries, and the validation steps taken. If any wording is reproduced, mark it as assisted content and cite appropriately.

• Verify everything produced with generative AI. At least two team members must review AI assisted code or text, rerun code end to end, and confirm correctness, fairness, and relevance to the business decision.

• Protect data and confidentiality. Do not upload the churn dataset, university materials, or any sensitive content to unapproved services. If you need to discuss data with a tool, use redacted or synthetic examples only.

• Keep the work reproducible. Any AI assisted code must run locally from a clean start with a fixed random seed and recorded package versions. Note AI assisted files in commit messages.

• Use real sources that are citations with verified, citable references. Do not rely on AI summaries without checking the original source.

• Maintain fairness and ethics. Do not use generative AI to bypass your fairness checks. You must explain model drivers and report group level performance using your own analysis.

• Be viva ready. Any team member may be asked to explain prompts used, code produced, model choices, and the business implications in a short walkthrough.

• Non compliance or undeclared use will be treated as academic misconduct and may affect marks under scholarly practice, communication, and integrity.