Prerequisites
Participants should have prior experience in Python programming. Familiarity with basic linear algebra is beneficial.
Participants should also be comfortable using AI-assisted tools to support coding and analysis.
Components
There are two parts in the Practice Module.
1. Practice Project: Participants will need to undertake at least one practice project to gain practical experience and demonstrate their understanding and mastery of the skills taught in the four component courses. The practice project will require an estimated 10 man-days of effort, which may be spread across several weeks or months. Projects may be conducted individually or in teams, depending on the nature of the requirements. Participants are expected to understand business needs for real-world projects, identify relevant data sources, build predictive models, and apply key analytical techniques to derive insights and solutions.
In the practice project, the module may incorporate complementary forms of assessment, such as scenario-based analytics exercises, verbal explanation of model outputs, and troubleshooting of analytical workflows—to evaluate participants’ critical thinking and practical problem-solving skills. These activities are designed to be lightweight yet effective in reinforcing real-world analytical competencies.
2. Examination: Each participant is required to sit for an examination on a stipulated date and time.
The overall grade for the participant will be based on the Practice Project and Examination.
Typical examples of projects to be undertaken
1. Problem description:
In this project, participants will develop and deploy an AI-powered customer churn prediction system for a fictional telecom company. The objective is to identify customers who are likely to leave the service in the next 30 days and provide interpretable insights that support proactive retention strategies.
Participants will start by working with structured customer data, such as account activity, billing history, and service usage patterns. They will also integrate unstructured data sources like customer service call transcripts or support ticket notes using natural language processing techniques. This multimodal dataset will be cleaned, transformed, and merged into a unified format using best practices in data preprocessing.
Predictive models — including logistic regression, decision trees, and ensemble methods like XGBoost or LightGBM — will be developed to classify churn risk. Participants will apply feature selection, dimensionality reduction, and hyperparameter tuning to optimise model performance.
One core component of the project is explainability. Participants will implement model interpretation techniques (e.g., SHAP or LIME) to generate stakeholder-facing reports that explain why specific customers are likely to churn. These explanations will form the basis for business decisions and interventions.
To complete the project, participants will build a data pipeline for automated scoring and deploy the model using an MLOps framework. Real-time monitoring, version control, and performance tracking will ensure the solution remains trusted and scalable in production.
2. Problem description
In this project, participants will design an AI solution for forecasting product demand in a Fast-Moving Consumer Goods (FMCG) sector. The task is to predict the weekly sales of various products based on historical sales data, product characteristics, customer reviews, and visual appearance.
Participants will begin by cleaning and transforming structured tabular datasets, including sales records, pricing history, and promotional campaigns. They will also extract and engineer features from unstructured data, such as product images (e.g. brightness) and customer review text (e.g. sentiment scores, keyword frequency), using image processing and natural language processing techniques.
Dimensionality reduction (e.g., PCA) will be applied where necessary to reduce noise and improve model generalisation. Participants will train regression and Deep Neural Network models to forecast product demand, comparing performance across algorithms and time horizons.
To simulate a real-world environment, the solution will be built with a modular data pipeline capable of automatically ingesting new data, retraining models, and updating forecasts. Unstructured data and metadata will be managed using NoSQL databases, with particular attention to the integration of multiple data types.
In addition to the technical deliverables, this project will also incorporate lightweight assessments to strengthen real-world analytical competencies. Participants may be required to complete short scenario-based analytics exercises that test their ability to interpret business problems, verbally explain model outputs and their implications to non-technical stakeholders, and troubleshoot issues in data preprocessing or model performance. These complementary activities reinforce critical thinking, communication, and practical problem-solving skills without adding significant overhead to the module.
Participants will then operationalise their models using an MLOps approach, deploy them into a cloud-based or containerised environment, and integrate results into a business-facing dashboard. The final deliverable will include forecast, explainability tools, and scenario planning features for decision-makers.
3. Problem description:
This project challenges participants to develop a real-time fraud detection system for a digital payments platform operating across multiple regions. The goal is to identify suspicious transactions in real time and provide clear, interpretable justifications to auditors and compliance teams.
Participants may work with synthetic transactional data representing millions of payment events, each containing customer metadata, transaction details, and behavioural signals. They will preprocess and engineer features to capture anomalies, such as sudden changes in transaction frequency, location mismatches, or unusual device usage. Advanced preprocessing techniques for outlier detection, handling missing values, and class imbalance will be applied.
The core modelling approach will involve ensemble classifiers and anomaly detection methods, which participants will evaluate and tune for precision, recall, etc. Given the critical nature of fraud detection, participants will implement explainable AI techniques to ensure that flagged transactions can be reviewed and justified by human experts.
The architecture of the solution will follow a decentralised, domain-driven design. Participants will simulate a data mesh structure by organising data by regional business units and integrating them into a shared pipeline using a data fabric or hub approach. NoSQL databases will store event logs and customer metadata.
To operationalise the system, participants will build a production-ready data pipeline capable of continuous scoring and feedback. They will use an MLOps framework to deploy and monitor the fraud detection models, ensuring robustness, scalability, and transparency under live conditions.
Deliverables and success criteria for the problem statements:
· Apply the concept of regression and classification to real-world predictive problems
· Implement machine learning models taught within this certificate
· Interpret analytical results and demonstrate the insights from the results
· Demonstrate the ability to apply appropriate methods and strategies to deploy and build trust in AI-driven data science solutions.
· Demonstrate thinking, communication and problem-solving skill