Machine learning is a very efficient tool when it comes to recognizing patterns in large datasets and making predictions by means of such patterns. Two general Machine learning methods are supervised and unsupervised learning. We further briefly describe machine learning methods and discuss how they could be adopted to solve certain telematics challenges.

A short introduction to machine learning

Machine learning could generally deal with both numeric and non-numeric problems, and is an iterative process by nature, meaning that it can not be completed in one go. The common routine of implementing a machine learning solution includes the following actions:

  • Selection of appropriate machine learning model and suitable algorithms
  • Training machine learning algorithm by means of training data
  • Building and testing the machine learning model
  • Results evaluation
  • Continue until the evaluation results matching the goal and complete the proposed model

Machine learning methods demonstrated reliable results in many applications, including pattern recognition (face detection), unusual patterns detection (sensors), and predictive analytics (stock/exchange rates and prices). Progress in machine learning largely relies on the advances in computations efficiency, general artificial intelligence progress, and artificial neural networks.

We discussed some bits on neural networks before, and here just briefly mention them. In neural networks, input units receive different forms of information that the actual neural network will try to learn about. Output units, located on the opposite side of the network, signal how it responds to the learned information. Hidden units are located in between input and output ones and form the major part of an “artificial brain”. The connections between units are represented by a weight, which is either a positive or negative number. The higher weight corresponds to a higher influence that one unit has on the other.

Supervised Vs Unsupervised learning

Most machine learning algorithms could be categorized based on the learner type. More specifically, there are generally supervised, unsupervised, reinforcement, and semi-supervised learning. Here we briefly mention supervised and unsupervised learning.

In a supervised learning scenario, the model provided with enough information/knowledge and essentially supervised during the learning process. Afterward, relying on the gained during the learning”experience”, it can try to predict the possible outcome of a new set of data. Depending on the model, the output may include discrete or continuous datasets. Supervised learning algorithms family includes decision trees, logistic regression, and Naive Bayes among the others.

Unsupervised learning case does not include any supervision routine performed to make a machine learning model learn. Instead, the actual model learns by itself relying on the data provided to it, and on its turn provides users with the details on patterns it managed to recognize during the analysis procedure. Typical examples of unsupervised learning are clustering and association rule learning algorithms.

Machine learning in telematics

Machine learning opens various fruitful opportunities for vehicle telematics. Taking into account current progress in general connectivity, IoT, and IoV, machine learning could actually utilize both real-time and historical data for insightful analytics. As such, machine learning methods could be employed in predictive maintenance, driver behavior analysis, insurance telematics, driver identification, ADAS systems, and autonomous vehicles.

To provide an example, let’s consider a Driver Identification system framework, described in [A. Girma et al., 2019].

The model developed by the authors is based on freely available vehicle telematics data being collected from the OBD-II interface of vehicles. The actual problem is formulated as a time-series prediction task, where the model is trained on sequential data obtained from an in-vehicle sensor. The approach proposed by the author’s claims to maintain its accuracy above the acceptable value, 88%, while other models’ accuracy goes below 40%.

Machine learning benefits from proven efficiency in solving various tasks including patterns recognition, predictive analytics, and dealing with huge and complex datasets. However, machine learning implementation could also be rather challenging, due to requirements on proper choice of algorithm, often lack clarity between training and test set, features engineering importance, and requirements on problem definition. Nevertheless, it is still a great option to extract useful and crucial information from enormous data volumes.

References

  • Y. Zhang. Machine Learning. 2010.
  • R. Prytz. Machine learning methods for vehicle predictive maintenance using off-board and on-board data. 2014.
  • R. Gopalakrishnan and A. Venkateswarlu. Machine Learning for Mobile. 2018.
  • H. Ahmadan, M. Jilani. Machine Learning for Automobile Driver Identification Using Telematics Data. 2019.
  • James Le. A Gentle Introduction to Neural Networks for Machine Learning. 2018.
  • A. Girma et al. Driver Identification Based on Vehicle Telematics Data using LSTM-Recurrent Neural Network. 2019.
  • https://www.navixy.com