| Titre : | The use of machine learning techniques for mobility prediction in vehicular networks | | Titre original : | L'utilisation des techniques d'apprentissages automatique pour La prédiction de mobilité dans les réseaux véhiculaires | | Type de document : | texte manuscrit | | Auteurs : | Hanane Amirat, Auteur ; Nasreddine Lagraa, Directeur de thèse | | Editeur : | Laghouat : Université Amar Telidji - Département d'informatique | | Année de publication : | 2020 | | Importance : | 121 p | | Format : | 27 cm | | Accompagnement : | 1 disque optique numérique (CD-ROM) | | Langues : | Anglais | | Catégories : | THESES :10 informatique
| | Mots-clés : | Real-time Route prediction Dependency graph Compact Prediction Tree Noise tolerance Lossless model POI recommendation Sequential rules mining | | Résumé : | The rapid development of location acquisition and mobile communication technologies have fostered a number of location-based services, such as location prediction and recommendation. While location prediction consists of forecasting the future location of a user or a moving object (e.g. vehicle), location recommendation aims at suggesting the next venues that a user may be interested to visit in the future. These services have several important applications in traffic congestion forecasting, location-based routing protocol designs, and targeting advertisements generation. In literature, various techniques and models have been proposed for effective location prediction and recommendation, e.g., Markov model, collaborative filtering. Although these models were shown to perform well, they suffer from several limitations and drawbacks. Among these we can mention information loss in prediction, noise sensitivity toward mobility data and static modeling in the recommendation. Besides, and due to the specificity of human mobility, the nature of mobility data and the continuous evolution of these services, the application of these methods in predicting and recommending locations become in many cases irrelevant and obsolete.
To deal with these issues, we propose, in this thesis, three models for location prediction and recommendation. The first proposed model is named NextRoute. This predictor is a novel and accurate lossless model as it compresses location data in a prediction tree without information loss, and it is designed to use all the relevant information contained in the training data to perform prediction. The second proposed model is MyRoute, a dependency-graph based predictor for real-time route prediction. MyRoute represents routes as a graph, which is then used to accurately match road network architecture with realworld vehicle movements. Unlike many prediction models, the designed model is noise-tolerant, and can thus provide high accuracy even with data that contains noise and inaccuracies such as GPS mobility data.
The third model we propose, in this thesis, is a rule based point of interest recommendation system we named STS-Rec. In addition to the sequential behavior of human mobility, this model takes also into account social and temporal influences. STS-Rec first transforms mobility data into location sequences. Then, it mines sequential recommendation rules from these sequences. The experimental evaluations we conducted on large-scale and realistic datasets show that our proposed models outperform several stateof-the-art models in terms of accuracy and coverage. | | note de thèses : | Thèse de doctorat en informatique |
The use of machine learning techniques for mobility prediction in vehicular networks = L'utilisation des techniques d'apprentissages automatique pour La prédiction de mobilité dans les réseaux véhiculaires [texte manuscrit] / Hanane Amirat, Auteur ; Nasreddine Lagraa, Directeur de thèse . - Laghouat : Université Amar Telidji - Département d'informatique, 2020 . - 121 p ; 27 cm + 1 disque optique numérique (CD-ROM). Langues : Anglais | Catégories : | THESES :10 informatique
| | Mots-clés : | Real-time Route prediction Dependency graph Compact Prediction Tree Noise tolerance Lossless model POI recommendation Sequential rules mining | | Résumé : | The rapid development of location acquisition and mobile communication technologies have fostered a number of location-based services, such as location prediction and recommendation. While location prediction consists of forecasting the future location of a user or a moving object (e.g. vehicle), location recommendation aims at suggesting the next venues that a user may be interested to visit in the future. These services have several important applications in traffic congestion forecasting, location-based routing protocol designs, and targeting advertisements generation. In literature, various techniques and models have been proposed for effective location prediction and recommendation, e.g., Markov model, collaborative filtering. Although these models were shown to perform well, they suffer from several limitations and drawbacks. Among these we can mention information loss in prediction, noise sensitivity toward mobility data and static modeling in the recommendation. Besides, and due to the specificity of human mobility, the nature of mobility data and the continuous evolution of these services, the application of these methods in predicting and recommending locations become in many cases irrelevant and obsolete.
To deal with these issues, we propose, in this thesis, three models for location prediction and recommendation. The first proposed model is named NextRoute. This predictor is a novel and accurate lossless model as it compresses location data in a prediction tree without information loss, and it is designed to use all the relevant information contained in the training data to perform prediction. The second proposed model is MyRoute, a dependency-graph based predictor for real-time route prediction. MyRoute represents routes as a graph, which is then used to accurately match road network architecture with realworld vehicle movements. Unlike many prediction models, the designed model is noise-tolerant, and can thus provide high accuracy even with data that contains noise and inaccuracies such as GPS mobility data.
The third model we propose, in this thesis, is a rule based point of interest recommendation system we named STS-Rec. In addition to the sequential behavior of human mobility, this model takes also into account social and temporal influences. STS-Rec first transforms mobility data into location sequences. Then, it mines sequential recommendation rules from these sequences. The experimental evaluations we conducted on large-scale and realistic datasets show that our proposed models outperform several stateof-the-art models in terms of accuracy and coverage. | | note de thèses : | Thèse de doctorat en informatique |
|
Ajouter le résultat dans votre panier Faire une suggestion Affiner la rechercheUne nouvelle approche basée sur les techniques de fouille de données pour la conception physique des entrepôts des données / Hanane Amirat
