الفهرس 
Machine Learning
MethodologyOnly 14 pages are availabe for public view
 |  | from | 92 |  |  |
| | | from | 92 | | |
Abstract
Driving on unfamiliar poorly paved roads is risky even if the vehicle speed is kept under limits. A driver may lose control if his vehicle suddenly comes into road anomalies, especially at night. In developing countries, road anomalies are not only common, but also new ones usually exist without precautions. We present an approach to the driver that detects and localizes road ruts by using no additional devices but hismobile phone. Our approach collects labeled data from mobile phones built-in sensors that describe road ruts. We use this data to feed a machine learning engine to build models that can detect new ruts. Our approach localizes identified ruts on the map via GPS coordinates and alerts drivers when they approach a rutted road. Our experiments show that the accuracy of the approach can be raised from 59% up to 99% if the learning technique is carefully selected and the sensors dataset size is increased to 100000 samples. In this thesis: Chapter 1, includes an overview on mobile phone sensors, road surface conditions, the problem statement, thesis contribution and the sensors that related to the rutted road-segment alert system. Chapter 2 includes a background of machine learning with mentioning classification algorithms that are fundamental for the rest work. The literature review is also discussed with the concentration on road surface conditions, outdoor localization and activity recognition fields. Chapter 3 includes rutted road segment alert system architecture that composed of five main stages that are: Collection of raw sensory data, filtering of sensory data, features extraction, classification, and localization. Chapter 4 includes the experiment setup and results discussion of the rutted road-segment alert system by applying a set of classifiers as logistic regression, naive Bayes and random forest. Chapter 5: includes conclusion and future work of the rutted road-segment alert approach. Finally the list of references followed by the appendix and the Arabic summary.