Machine learning models are powerful tools used to efficiently and effectively perform vital tasks and solve complex problems. An exponential increase in data across the modern world means organisations from a range of sectors are ready to deploy machine learning models. These models have a huge range of uses, whether machine learning in finance proactively monitoring bank transfers for signs of fraud, or machine learning in healthcare powering the next generation of diagnostic tools.
The process of building a machine learning model is often complex, driven by specialists in data science. But an understanding of the process is important as machine learning is adopted by more and more organisations. This guide explores the basics of building a machine learning model, breaking the process up into six steps.
Six steps to build a machine learning model
Although different types of machine learning will have different approaches to training the model, there are basic steps that are utilised by most models. Algorithms need large amounts of high quality data to be effectively trained. Many of the steps deal with the preparation of this data, so that the model can be as effective as possible. The whole project needs to be properly planned and managed from the beginning, so that a model fits the organisation’s specific requirements. So the initial step deals with contextualising the project within the organisation as a whole.
The six steps to building a machine learning model include:
- Contextualise machine learning in your organisation
- Explore the data and choose the type of algorithm
- Prepare and clean the dataset
- Split the prepared dataset and perform cross validation
- Perform machine learning optimisation
- Deploy the model
Contextualise machine learning in your organisation
The initial step in building a machine learning model is to understand the need for it in your organisation. The machine learning development process can be resource intensive, so clear objectives should be agreed and set at the start. Clearly define the problem that a model needs to solve and what success looks like. A deployed model will bring much more value if it’s fully aligned with the objectives of your organisation. Before the project begins, there are key elements that need to be explored and planned.
At this stage the following details should be agreed:
- The overall owners of the machine learning project.
- The problem the project needs to solve, and a definition of project success.
- The type of problem the model will need to solve.
- The goals of the model to understand return on investment once deployed.
- The source of training data and whether it is of sufficient quantity and quality.
- Whether pre-trained models can be deployed instead.
If a pre-trained model can be realigned and reused to solve the given problem, the process of building a machine learning model will be streamlined. Instead of building a model from scratch, the process of transfer learning can reuse an existing model to solve a similar problem. This will cut down on the resources required for the project, especially in supervised learning which requires large arrays of labeled training data.
Explore the data and choose the type of algorithm
The next step in building a machine learning model is to identify the type of model that is required. The differences depend on the type of task the model needs to perform and the features of the dataset at hand. Initially the data should be explored by a data scientist through the process of exploratory data analysis. This gives the data scientist an initial understanding of the dataset, including its features and components, as well as basic grouping.
The type of machine learning algorithm chosen will be based on an understanding of the core data and problem that needs to be solved. Machine learning models are broadly split into three major types. Each one has a distinct approach to training the model. Supervised machine learning models require labeled datasets which have been prepared by a data scientist. The training dataset will therefore include input and labeled output data. The model then learns the relationship between input and output data. Supervised machine learning models are used to predict outcomes and classify new data.
Unsupervised machine learning models are trained on unlabeled datasets. The training dataset just requires input variables. This type of machine learning model learns from the dataset, and is used to identify trends, groupings or patterns within a dataset. This type is mainly used to cluster and categorise data, and detect the rules that govern the dataset.
The third main type of machine learning algorithm is the technique of reinforcement machine learning. This process takes a trial and error or feedback loop approach to learning. Reward signals are released whenever a successful action is performed, and the system learns through trial and error. An example of reinforcement learning is in the development of driverless cars. Systems learn through interacting with the environment to perform a specific task, learning and improving from its past experiences.
Prepare and clean the dataset
Machine learning models generally need large arrays of high quality training data to ensure an accurate model. Generally, the model will learn the relationships between input and output data from this training dataset. The makeup of these datasets will differ depending on the type of machine learning training being performed. Supervised machine learning models are trained on labeled datasets, which contain both input variables and labeled output variables.
The process of preparing and labeling the data is usually completed by a data scientist and is often labour intensive. Unsupervised machine learning models on the other hand won’t need labeled data, so the training dataset will just contain input variables or features. In both types of machine learning the quality of data has a major effect on the overall effectiveness of the model. The model learns from the data so poor quality training data quality may mean the model is ineffective once deployed. The data should be checked and cleaned so data is standardised, any missing data is identified, and any outliers are detected.
Split the prepared dataset and perform cross validation
The real-world effectiveness of a machine learning model depends on its ability to generalise, to apply the logic learned from training data to new and unseen data. Models are often at risk of being overfitted to the training data, which means the algorithm is too closely aligned to the original training data. The result will be a drop in accuracy or even a loss in function when encountering new data in a live environment.
To counter this, the prepared data is usually split into training and testing data. The majority of the dataset is reserved as training data (for example around 80% of the overall dataset), and a subset of testing data is also created. The model can then be trained and built off the training data, before being measured on the testing data. The testing data acts as new and unseen data, allowing the model to be assessed for accuracy and levels of generalisation.
The process is called cross validation in machine learning, as it validates the effectiveness of the model against unseen data. There are a range of cross validation techniques, categorised as either exhaustive and non-exhaustive approaches. Exhaustive cross validation techniques will test all combinations and iterations of a training and testing dataset. Non-exhaustive cross validation techniques will create a randomised partition of training and testing subsets. The exhaustive approach will provide more in-depth insight into the dataset, but will take much more time and resources in contrast to a non-exhaustive approach.
Perform machine learning optimisation
Model optimisation is an integral part of achieving accuracy in a live environment when building a machine learning model. The aim is to tweak model configuration to improve accuracy and efficiency. Models can also be optimised to fit specific goals, tasks, or use cases. Machine learning models will have a degree of error, and optimisation is the process of lowering this degree.
The process of machine learning optimisation involves the assessment and reconfiguration of model hyperparameters, which are model configurations set by the data scientist. Hyperparameters aren’t learned or developed by the model through machine learning. Instead, these are configurations chosen and set by the designer of the model. Examples of hyperparameters include the structure of the model, the learning rate, or the number of clusters a model should categorise data into. The model will perform its tasks more effectively after optimisation of the hyperparameters.
Historically, the process of hyperparameter optimisation may have been performed through trial and error. This would be extremely time consuming and resource intensive. Now, optimisation algorithms are used to rapidly assess hyperparameter configuration to identify the most effective settings. Examples include bayesian optimisation, which takes a sequential approach to hyperparameter analysis. It takes into account hyperparameter’s effect on the target functions, so focuses on optimising the configuration to bring the most benefit.
Deploy the machine learning model
The last step in building a machine learning model is the deployment of the model. Machine learning models are generally developed and tested in a local or offline environment using training and testing datasets. Deployment is when the model is moved into a live environment, dealing with new and unseen data. This is the point that the model starts to bring a return on investment to the organisation, as it is performing the task it was trained to do with live data.
More and more organisations are leveraging containerisation as a tool for machine learning deployment. Containers are a popular environment for deploying machine learning models as the approach makes updating or deploying different parts of the model more straightforward. As well as providing a consistent environment for a model to function, containers are also intrinsically scalable. Open-source platforms like Kubernetes are used to manage and orchestrate containers, and automate elements of container management like scheduling and scaling.
Deploy machine learning models with Seldon
Products such as Seldon Deploy are used to streamline model deployment and management. It’s a language-agnostic platform which integrates a deployed model with other apps through API connections. It also provides workflow management tools for a streamlined deployment process, and an analytics dashboard to monitor the health of the model. Once deployed, it’s vital that the model is continuously monitored for model drift so that it stays accurate and effective.
Seldon Technologies will help your organisation serve machine learning models. Since 2014, Seldon Technologies has been working to democratise access to machine learning. Take a step towards embedding machine learning in your organisation with help from Seldon.
Deploy machine learning in your organisations effectively and efficiently. Talk to our team about machine learning solutions today.
Graeme is VP of Marketing at Seldon, for the past 10 years he has led the marketing function in AI and Infrastructure vendors across EMEA and the US. Graeme has contributed to a number of industry ebooks and spoken at various events. Outside of work he enjoys family time, cycling and golf.