Applying AI-algorithm to predict clusters the road traffic data

Abstract

Forum Virium Helsinki and Flow Analytics have had a project together where they have installed 3 LiDAR sensors in Helsinki Esplanade area to collect data in the traffic. Dataset contains 9 columns where 1 of them is label. The label column includes car, pedestrian and cyclist groups. The problem is that dataset doesn’t provide enough of information. They would like to use data to improve area’s safety and attractiveness however car label doesn’t recognise is it either light or heavy vehicle. There also are the electric scooters in the traffic however they haven’t been recognised in dataset.

The main objective of this thesis was to investigate whether AI is capable of identify a new class within the dataset and reclassifying car labels as light or heavy. To achieve this objective, unsupervised learning was applied using clustering techniques such as k-means, Gaussian Mixture Models (GMM) and autoencoder. K-means made some mistakes and results weren’t so clear than with autoencoder however good enough. Nevertheless, autoencoder and k-means were able to predict clusters as expected.