IBM/Presidion follows a structured process known as CRISP-DM. As part of this process, there will be clear milestones that will allow NGN and IBM/Presidion to monitor project progress and success to date. This process has 6 high level stages:
- Business Understanding: Typically, all interested and relevant parties are engaged to fully scope and define the business problem and resources available. A project plan will result from these meetings that will outline specific tasks and milestones and what resources will be needed for each.
- Data Understanding: At this stage, the IT team are initially involved so that Presidion can best understand the processes behind data collection. An assessment will be made as to the quality of the data held, and what is initially considered as likely factors in affecting failure to meet the 7/28 target. If the relevant data cannot be readily accessed, the project will be immediately halted at this point, and only resumed when the data access issue is fully resolved.
- Data Preparation: As the data for the project comes from a variety of different sources, and typically these are recorded at different levels, the information will be required to be manipulated and collated in a manner efficient for predictive modelling.
- Modelling: Using IBM-SPSS Modeller, Presidion will build predictive models to associated risk scores with cases. These risk scores will identify the major factors associated with a case failing to meet both the 7 and 28 day targets. The score itself, can act as a scientific tool for resource allocation and deployment strategies.
- Evaluation: Before deployment of the predictive models, Presidion will evaluate their accuracy, understanding and usability. This may result in alternative models being produced. Expected levels of prediction can then be used to inform NGN of likely successful deployment strategies.
- Deployment: As part of the deployment, Presidion will interface with existing NGN processes and systems to best get the score information to the Area Managers. Additionally, documentation will be provided to describe the process behind building the models and what main factors were identified.
Benefits
The challenge is to test if Predictive Analytics can outline, using available NGN data, the factors associated with reported gas leaks and predict where the next geographic location for a leak is most likely to be. Would such a model enable more effective deployment of the resources by the Area managers, and give valuable additional insights towards planning and decision making. Success criteria for the Predictive Maintenance project have been identified:
- Accurate/repeatable identification of when and where a gas leak is most likely to be reported
- Understanding of the main factors affecting a case
- Demonstration of the benefit of deploying predictive modelling to business areas to enable more effective fleet/resource planning
- Capability to take ownership of the data models created, as part of the NGN roll out of Predictive Analytics.