Level Crossings are among the most critical railway assets, concerning both the risk of accidents and their maintainability, due to intersections with promiscuous traffic and difficulties in remotely monitoring their health status. Failures can be originated from several factors, including malfunctions in the bar mechanisms and warning devices, such as light signals and bells.
One of the problems of such an approach is in the high variability of the transformation efficiency, which is strongly affected by factors associated with the characteristics of the source material, pre-processing stages, storage type, manipulation etc. The aim of the second Machine Learning Mini-Contest (MC2) for the academic year 2021/2022 is to predict the Oxygen/Carbon ratio (numeric prediction) for a given raw feedstock sample described by its properties and characteristics.
Among all, the detection of the level crossing alarm bell is an important task, as it represents one of the most effective notifications for vehicles' drivers of an upcoming train on the track. Unfortunately, its detection is not straightforward, due to i) the presence of other similar sounds (e.g. alarms), ii) the traffic noise and iii) the great variability of alarm bell sounds. The aim of the third Machine Learning Mini-Contest (MC3) for the academic year 2021/2022 is to detect whether a given sound sample belongs to a warning bell at level crossings or not (multi-class classification).
Each student has to predict (multi-class categorical classification) the source of the sound provided as a sample, realising one or more prediction models using data analysis and Machine Learning techniques. The performance measure to maximise is Accuracy. It is mandatory for the student who will achieve the best performance on the test dataset, to discuss the process steps followed in order to reach the development of the final model. The winning student presentation will be held during the lesson on December the 17th. If the presentation and the proposed solution will be judged positively, the author will be relieved from discussing one of the contest solutions during the final exam. Each participant is free to use external tools (i.e. Weka, Knime, MatLab, etc.).
This task is part of the H2020 Shift2Rail project "RAILS: Roadmaps for A.I. Integration in the Rail Sector". We thanks prof. Valeria Vittorini and prof. Francesco Flammini for providing the data.
Kaggle: https://www.kaggle.com/c/unina-machine-learning-2122-minicontest-n3