Evaluating Incremental Machine Learning Models for Road Condition Classification
| dc.contributor.author | Svantesson, David | |
| dc.contributor.author | Hansen, Julia | |
| dc.contributor.department | Chalmers tekniska högskola / Institutionen för data och informationsteknik | sv |
| dc.contributor.department | Chalmers University of Technology / Department of Computer Science and Engineering | en |
| dc.contributor.examiner | Duregård, Jonas | |
| dc.contributor.supervisor | Breitholtz, Adam | |
| dc.date.accessioned | 2024-09-10T08:13:54Z | |
| dc.date.available | 2024-09-10T08:13:54Z | |
| dc.date.issued | 2024 | |
| dc.date.submitted | ||
| dc.description.abstract | This project was provided by Klimator AB with the aim of evaluating different incremental Machine Learning (ML) models predicting Road Surface Conditions (RSC), using a given data set from Klimator AB. Successfully identifying the RSC is associated with an autonomous vehicle’s traffic safety and therefore an important area of investigation. This project presents the evaluation of seven models, with three of the models, Gaussian Naive Bayes, Complement Naive Bayes and Hoeffding Tree Classifier, being of an incremental nature in their implementation, and the remaining, Decision Tree Classifier, K-Nearest Neighbors, Logistic Regression and Dummy Classifier, employed as ensembles in an incremental manner. The models were evaluated against a Random Forest model serving as a top-level baseline, and a single Dummy Classifier serving as a low-level baseline. All models were trained using datasets derived from vehicle rides under varying RSCs, which were split into smaller intervals called epochs. The findings of this study are that the ensembles of Logistic Regression and Decision Tree Classifier demonstrate the greatest overall strengths, achieving the highest average accuracy. Additionally, the Hoeffding Tree Classifier performs strongest during rapid changes of RSC, so-called concept driftevents. However, the performances of all models fall short of optimal in terms of making accurate predictions. To optimize the top three candidates further, this project identifies opportunities for further development and enhancements, potentially leading to an ML model suited to assist in autonomous vehicles and ensuring traffic safety. | |
| dc.identifier.coursecode | LMTX38 | |
| dc.identifier.uri | http://hdl.handle.net/20.500.12380/308554 | |
| dc.language.iso | eng | |
| dc.setspec.uppsok | Technology | |
| dc.subject | Machine Learning | |
| dc.subject | Incremental Learning | |
| dc.subject | Road Surface Conditions | |
| dc.subject | Ensemble Method | |
| dc.subject | Concept Drift | |
| dc.subject | Traffic Safety | |
| dc.subject | Model Evaluation | |
| dc.subject | Autonomous Vehicles | |
| dc.title | Evaluating Incremental Machine Learning Models for Road Condition Classification | |
| dc.type.degree | Examensarbete på grundnivå | sv |
| dc.type.uppsok | M | |
| local.programme | Datateknik 180 hp (högskoleingenjör) |