The use of machine learning (ML) for solving day-to-day business problems is gaining steam. A variety of ML approaches, such as supervised learning, unsupervised learning, and reinforcement learning, are used to design and build solutions in this regard. These approaches involve use of classification, regression, and clustering based algorithms, to name a few, to make predictions about a matter of interest.

As the term suggests, a classification-based algorithm will help predict a class or category, or, in other words, make predictions about a discrete variable of interest, such as a successful or not successful project. Whereas a regression-based algorithm will be able to predict a continuous variable such as the life cycle cost of a project. For the sake of clarity, a continuous variable can take infinite values compared to a discrete variable that can take finite values. The understanding of this distinction is vital when choosing a solution to a given problem in any domain, and project management is no exception. Clustering based algorithms, on the other hand, are used when we want a program to group data based on inherent characteristics of the data as data labels are not available, such as when a project has not been categorized as a successful or not successful project.

While all types of algorithms can be used for developing solutions in project management and certainly there are no restrictions, for the sake of simplicity, here we will look at the pros and cons of using classification vs. regression-based algorithms and try to develop an understanding if either of the two is more suited to project management given the nature of project-based work.

Classification-based algorithms

The nature of project management often means that the problem that we handle entails predicting a class or a category. This can happen for a number of things:

1) Resource assignments
The success of projects is largely dependent on people, i.e., their skills, attitudes, and professionalism. As such, identifying the right people can be the only difference between a failed vs. successful project. In particular for projects that involve large budgets and high stakes, choosing the right people becomes even more critical. A classification-based algorithm can help in predicting the level of staff (junior vs. senior, number of years of experience) that could be used for a particular project to ensure the best possible outcome.

The same thought process can be extrapolated when deciding the team assignments. That is to predict which team should be assigned to which project. The teams can be labelled based on their level of experience, past performance, project success contributions, etc., for the system to learn from data of what attributes contribute to which type of project success to predict the team that would be most suitable given a particular project.

2) Risk identification
Classification-based algorithms will also be suited to predict the class or category of risk associated with a project. We can use classification-based algorithms to predict the severity (e.g., high, medium, or low) and likelihood of occurrence (e.g.,high, moderate, or negligible) category.

3) Project success or failure
The projects involve significant investments of time and money. So project investment decisions should be made based on sound rationale based on the available data. Classification-based algorithms can help in making decisions by predicting if the project will be a success or a failure based on project attributes such as the level of project staff experience, risk management effectiveness, project manager skills, and project governance, to name a few.

The predictions made using classification-based algorithms can be critical, particularly for projects with large capital investments and significant time involvement.

Regression-based algorithms

1) Project timeline
Predicting project timeline or project completion date could be done using regression- based algorithms. The attributes such as task duration, dependencies, resource allocations, and risk-related data about task getting delayed could be useful for regression
analysis and prediction.

2) Effort estimates
Regression-based algorithms will also be suited to estimating the effort needed to complete the project or certain tasks. Prediction of effort can then be used for duration estimates and scheduling of tasks. Needless to mention, understanding could help reduce costs to significant levels. Often, a lack of understanding of the efforts involved in completing a task leads to over- or underestimation of task duration and hence sequences of tasks.

Concluding thoughts:

The evolving role of ML in solving business and day-to-day problems necessitates building an understanding of how it can be used effectively. The use of it for effectively managing projects cannot be overstated, as project delivery is an ongoing process that happens across every business and industry sector.

Given such an importance, we have looked at the suitability of classification vs. regression-based algorithms for different scenarios in a supervised learning context. To work on classification-related problems, one can use algorithms such as decision trees and logistic regression, and for regression-related problems, one can use linear regression and polynomial regression, among others. 

Certainly, the discussion in this article is limited. The intent is to develop a discourse, and the article is by no means conclusive or exhaustive in any way. More work is surely needed to explore how classification vs. regression-based algorithms can be used. One thing is for sure: no one particular algorithm is better or worse. In some cases, using both classification and regression-based algorithms for supervised learning is necessary to solve complex problems.

© 2024 Jiwat Ram, All Rights Reserved.