ANÁLISIS DEL COMPORTAMIENTO SEGURO/AGRESIVO DEL CONDUCTOR DESDE UNA PRESPECTIVA DE MACHINE LEARNING

JACINTO SIMÓN, ORALDO (2020) ANÁLISIS DEL COMPORTAMIENTO SEGURO/AGRESIVO DEL CONDUCTOR DESDE UNA PRESPECTIVA DE MACHINE LEARNING. Maestría thesis, Universidad Autónoma de Chihuahua.

[img]
Vista Previa
Text
Tesis_FING_Oraldo(29-1-2020).pdf

Download (3717Kb) | Vista Previa

Resumen

Aggressive car driving is known as the driver's behavior that directly or indirectly attempts to provoke, harm or cause damage to any type of other people who share the common space through which traffic on the road runs. The analysis of driving behavior through the use of mobile technology has shown that driving safety can infer data collected by Smart Phone (IT) sensors. The analysis of driving behavior through the use of mobile technology has shown that driving safety can be inferred from the data collected by smartphone sensors. The machine learning classification algorithms on the data of a smartphone are widely used in the classification of maneuvers to detect aggressiveness and it is demonstrated that they can accurately detect aggressive events such as braking, acceleration, turns and lane change. The effectiveness of a machine learning classification algorithm for identifying aggressive driving events depends largely on the quality of the data and the method of extracting features for its representation. There are several ways of representing the data that in combination with classification algorithms allow identifying aggressive maneuvers efficiently. The best papers reported in the literature use different forms of representation and do not use the same metrics to measure their results. These results are not comparable and it can be seen that not enough has been done to determine the best solution to this problem. The objective of this investigation is to evaluate the safe/aggressive behavior of the driver of a vehicle through smartphone data. For this, it is experimented with the methods of extraction and representation of characteristics of better results reported in the literature: (representation in statistical values, Bag of Word (BoW) and the minimum rules algorithm MODLEM). The best combination parameters plus classifier according to the results obtained in the Area Under Curve (AUC) metric in 8 proposed classifiers. The best results were obtained: • The MODLEM algorithm together with Fisher Linear Discriminant Analysis (ADLF) obtained 100% in the classification activity in a data set of 69 aggressive events but it did not obtain good results in the rest of the experiments with other data sets. • The increase in the size nf of sliding windows in the representation of statistical values improves the classification activity, the MHLDForest classifier being the one with the best results with an AUC of 99.3%. • With the use of ADLF on the representation of BoW, the classification activity improves, with the MLP classifier being the one with the best results with an AUC of 99.9%.

Tipo de Documento: Tesis (Maestría)
Palabras Clave: aggressive driving, smartphones, sensors, machine learning
Clasificación temática: Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Usuario Remitente: Admin Administrador del Respositorio
Depositado: 28 Jul 2020 18:31
Ultima Modificación: 28 Jul 2020 18:31
URI: http://repositorio.uach.mx/id/eprint/273

Actions (login required)

Ver Objeto Ver Objeto

Universidad Autonoma de Chihuahua