Skip to main content

Skills application

Instruction and application
In Progress

The optimisation diagnosis

In this skills application, you will investigate and resolve performance issues in a predictive model by diagnosing its training inefficiencies and applying optimisation and regularisation techniques to achieve better generalisation.

Skills application illustration

Context

Organisation: You have joined MedPredict, a health technology startup focused on developing machine learning models for early disease detection.

Current challenge: The team's latest neural network performs exceptionally well on the training dataset but fails to generalise to new, unseen patient data. This has raised concerns about potential overfitting and unstable convergence during training.

Your role: As the ML Engineer, your task is to investigate the issue, identify the causes of poor generalisation, and apply optimisation and regularisation strategies to improve model stability and performance.

Objective: By the end of this activity, you will have refined the model's training process to achieve more consistent and reliable predictive outcomes.

Success criteria

To complete this skills application successfully, you must:

  • Identify and explain the model's training and generalisation issues.
  • Compare optimisers and justify which performs best for this dataset.
  • Apply and evaluate regularisation techniques to reduce overfitting.
  • Recommend one improvement to enhance model stability or performance.
  • Reflect on the balance between convergence speed, generalisation, and efficiency.

Instructions and materials

Follow the steps below to complete this skills application. Completing this activity will unlock the solution example on the following page.

Download the activity resources

Select the links below to download the Jupyter Notebook and dataset for this activity. The notebook provides the step-by-step guidance, while the dataset file is required to complete the model training and optimisation tasks.

Access the Jupyter Notebook and dataset

  • Open the provided Jupyter Notebook for this activity and download the accompanying dataset.
  • Upload the dataset to your notebook environment before running the code.
  • The notebook contains all baseline code and step-by-step guidance to help you complete the model training and optimisation tasks.

Complete the tasks

Follow the instructions in the notebook to:

  • Compare optimisers and evaluate their effect on model convergence and generalisation.
  • Apply regularisation techniques to reduce overfitting and improve model stability.
  • Recommend one adjustment to enhance overall model performance.

Record and submit

Document your findings and reflections directly in the notebook. Save your completed file and submit it for review.