Good Machine Learning Practice (GMLP)

Here we outline and define the key components of Good Machine Learning Practice (GMLP) in the manufacture of medical devices.

What is GMLP?

Good Machine Learning Practice – or GMLP – is a set of guidelines, methodologies, and ethical standards that aim to guide the responsible development, deployment, and maintenance of machine learning systems.

1) Data Collection in GMLP

This is the process of gathering and measuring information on targeted variables in an established systematic fashion, which is crucial for creating reliable and accurate machine learning models. This involves ethical data sourcing and quality assurance:

Ethical Data Sourcing in GMLP

An essential part of GMLP is that data is gathered responsibly and legally, inline with the emerging guidelines and regulations to ensure integrity and trust.

• Anonymisation
• Consent
• Transparency

Data Quality in GMPL

• Cleaning
• Normalisation
• Outlier Detection

2) Model Development

Model Development is the systematic process of creating, training, and refining a machine learning model to make accurate predictions or decisions.

Feature Engineering

Feature engineering is the process of selecting, modifying, and transforming variables to create input data that improves model performance and accuracy.

• Dimensionality Reduction: Techniques used to optimise data for machine learning models by selecting relevant features and reducing the number of variables.
• Feature Scaling: Standardising the range of independent variables to ensure each feature contributes equally to the analysis.
• Feature Selection: Identifying and selecting the most relevant data attributes to improve model performance and efficiency in machine learning.

Model Training

Iterative process of improving model performance by adjusting algorithms and hyperparameters using labeled data.

• Cross Validation
• Hyperparameter Tuning
• Ensemble Methods

3) Model Evaluation

Model Evaluation assess the performance and generalisation ability of machine learning models to ensure their effectiveness in real-world applications:

Performance Metrics

Performance metrics are methods that are used to assess the quality and effectiveness of machine learning models in solving specific problems.

• Accuracy
• Precision
• Recall

Validation Strategies

Techniques used to assess the performance and generalizability of machine learning models.

• Train-Test Split
• K-Fold Cross Validation
• Leave-One-Out Cross Validation

4) Deployment

The process of making trained machine learning models available for use in a production environment.

Scalability

Ensuring models are accurate, reproducible, and interpretable, ready for use in real-world applications that require handling large datasets and high loads.

• Paralellisation
• Distributed Computing
• Containerisation

Monitoring

• Real-time monitoring
• Error analysis
• Bias Detection

5) Ethical Consideration

Fairness

Ensuring unbiased and equitable outcomes in machine learning models through fair treatment and consideration of ethical implications.

• Bias Mitigation
• Fairness Testing
• Algorithmic Transparency

Privacy Aspects

• Data Encryption
• User Consent
• Data minimisation