Knowledge check

Knowledge check

Complete

Evaluate your understanding of this unit

Answer the following questions on methodologies, risk and delivery.

You're managing a project to develop a fraud detection model using structured financial transaction data. The business team wants to stay involved throughout the process, and the organisation needs documentation and checkpoints for compliance. Which project management methodology is the best fit?

A) Agile/ScrumB) CRISP-ML(Q)C) WaterfallD) Kanban

You're building a model for internal use that classifies legal documents. To keep stakeholders informed and ensure alignment, which element should be included in your ML project plan?

A) A complex architecture diagram sent once at the beginning.B) A lightweight communication plan with cadences and responsibilities clearly defined.C) A real-time dashboard limited to the ML team.D) A checklist of technical tasks without linking them to business outcomes.

What is a common mistake when estimating computational resources for ML projects?

A. Underestimating the computing power required for model training.B. Always choosing cloud-based GPUs over on-premise solutions.C. Ignoring the need for distributed computing when handling large datasets.D. Assuming all ML models require the same level of compute power.

How can ML teams ensure models remain effective over time?

A. Continuously monitor model performance and retrain as needed.B. Deploy models once and assume they will work indefinitely.C. Remove old models without analysing performance changes.D. Only update models if an error occurs in production.

What is a major risk in ML projects related to data?

A. Data drift, where incoming data distributions shift over time.B. The model being too complex.C. A lack of explainability in model decisions.D. Using open-source tools instead of proprietary solutions.

What is a key strategy for balancing multiple ML projects in an organisation?

A. Prioritising projects based on business impact and feasibility.B. Assigning equal resources to all projects.C. Focusing only on experimental research projects.D. Ignoring low-priority projects altogether.

Why do ML projects require ongoing resource allocation beyond deployment?

A. ML models degrade over time and require updates to maintain performance.B. ML models run indefinitely without additional resources.C. Once deployed, ML models do not require monitoring.D. Retraining is only necessary if the model completely fails.

You're working on a predictive maintenance model for industrial machinery. The data is messy, the business problem is loosely defined and the team needs time to explore potential solutions. What approach best supports this type of project?

A) Use Agile sprints to deliver a fully working model within two weeks.B) Apply CRISP-ML(Q) to guide discovery and align with business needs before iterative model development.C) Use Waterfall to define all requirements up front.D) Assign separate workstreams to data, modelling and integration to accelerate progress.

What is the primary reason for version control in ML projects?

A. To track changes in data, models and code for reproducibility.B. To keep multiple copies of the same model.C. To avoid using open-source tools.D. To ensure ML models are always complex.

What is an effective way to ensure ethical considerations in ML projects?

A. Conducting fairness audits and bias testing.B. Ignoring regulatory requirements.C. Keeping AI decisions fully opaque.D. Relying solely on automation without human oversight.

In progress