Machine Learning Implementation for Project Management
Part of my research is focused on machine learning algorithms and how they can be used to improve the project methodology. Here is a practical example of how this works. There is a risk that occurs on some projects and not on other projects and the project manager wants to know the probability that this risk will occur during the new project. Data is captured in a database or spreadsheet as shown below with the rows representing each project dataset and the columns representing the project characteristics which are called features in machine learning terminology. The data in each column is represented as a 1, if the defined feature existed in the project, or 0, if the feature was not present in the project. There can be a large number of features depending on the available data and perceived value of each feature. At the end of each row is a label that identifies if the risk occurred, 1, or did not occur, 0, for that specific project. This is an example of supervised learning since each dataset is labeled.
A small number of lines of software code are used to create a model that represents the data. For this algorithm, I use a three-layer neural network which is a fancy way of saying a program that uses calculus equations to find the closest correlations in the data. For a new project, the data is collected for all of the features and that data is submitted to the trained model. The output is a probability that the identified risk will occur in the new project, based on the project characteristics. This example uses risk but the same algorithm can be used for a quality issue, a scope issue or any other factor than can affect a project, as long as the appropriate data is provided. The output can be taken a step further. With a risk that is likely to occur, for example, an automatic response can be made by an AI tool to change the project schedule to avoid the risk from occurring.
This illustrates two characteristics of machine learning: the critical importance of project data and the simple reusability of a basic machine learning algorithm. Of course, I have simplified the concept in an attempt to share knowledge about this new technology that is at the heart of advancements by industry giants like Amazon and Google. AI is available now and anyone who has a creative way to evaluate project issues can construct a model to take advantage of this technology. A virtual tsunami of AI tools for project management are on the way and all of them are based in some way on this concept.
Published at pmmagazine.net with the consent of Paul Boudreau