The Dutch public procurement market is a multi-billion industry, with a total value of 73 billion euros per year. Competitive tendering is the most popular method of selecting a supplier for the required construction services. The realisation of a tender bid is an expensive and complex process, established on the intersection of various disciplines e.g. safety, constructability, finance, cost estimation and risk management.

Machine Learning has been a popular method in various industries to predict future outcomes and uncover patterns in historical data but remains a rather novel phenomenon in the construction industry. Machine Learning models have been developed in the past to aid tender management, but not with a focus on predicting the contractor’s tender price.

The objective of this research is to develop a Machine Learning tool that is able to predict the tender price of infrastructure projects accurately and is able to assist the contractor’s tender professionals in their decision to tender.

In order to achieve this objective, the following research question was formulated:
How can a Machine Learning algorithm, predicting the tender’s price using tender project data, be developed to support the contractor’s decision to tender?

First, a literature study is conducted to explore the state-of-the-art developments of Machine Learning within construction tender management, discover what the most popular regression algorithms are and what the most important tender features are that influence the tender price. Based on the most important tender features and interviews with tender professionals of a Dutch contractor, an extensive list of 93 tender features is filtered until a final set of tender features remains. Data on these tender features are collected in order to be used as input for the Machine Learning model.

The SVR model performed the best with an R-Squared of 0.846, implying that 84.6% of the variance of the tender’s price could be explained by the model. The SVR model includes an optimised set of features, which is a subset of the initial dataset. The initial estimate is considered to be significantly more important than the other features.

In order to implement the Tender Price Predictor in the organization of a Dutch contractor, attention should be paid to the required effort of the users and to ensuring a high quality of tender data. Requiring too much effort from tender managers entering the input data into the database may result in worse quality of data. Both the users of the model and the managers submitting tender data should be trained accordingly in order to obtain maximum effectiveness.

It should be noted that some important features, according to practice and literature, are omitted from the final dataset. Although the final set of features complies with the requirements of generic features and sufficient occurrences in either literature or the interviews, they were not present in the database of the contractor. These omitted features are: ‘Project team experience’ and ‘Location’