Artificial neural networks could be described as a set of algorithms, simulating lots of brain cells inside a computer, that are designed to recognize patterns. In a simplified way, they could be represented as a layer performing clustering and classification, which is located on top of the stored/managed data. Here we will discuss how beneficial artificial neural networks could be to various telematics applications.

A quick intro to neural networks

We have already discussed some basics behind neural networks in our blog. Therefore, here we just briefly go through some aspects. Artificial neural networks are fundamentally a software simulation, including collections of algebraic variables and mathematical equations linking them together.

An artificial neuron to some extent mimicking the biological neuron and consists of a neuron body and connections to other neurons. Each input to neuron has associated weight, indicating the impact of a particular input to calculations. Neuron’s body has two components, each associated with input (represents a part of calculations that neuron performs) or output. A typical neural network consists of many artificial neurons called units. These units are arranged in sets of layers, each of which connects to the layers on either side.

A special type of neural network called convolutional neural networks. Compared to “regular” neural networks, convolutional neural networks are more advantageous for image/video/face recognition applications but have more complex architecture. 

There are two ways for information flow to propagate throw the network: forward network and backpropagation. Backpropagation helps the network to learn by decreasing the difference between intended and actual output. Weights are initialized randomly at a start and are learned as actual computational models provided with more and more data.

At the summation stage, the transposed feature matrix multiplied with a weight matrix to represent the collective influence of all input features and their respective weights by a single scalar value. Further introduced nonlinearity allows a neuron to model more complex real-world patterns. A common way to introduce such nonlinearity is to use so-called activation functions.

The output that the network has produced is compared with the output it is meant to produce, and the difference between them is applied to modify the weights of the connections.

Neural networks in telematics: scenarios

Neural networks are proved to be an efficient tool for solving clustering and classification tasks, including those rising in telematics. For instance, convolutional neural networks are successfully utilized to address various image/video/face recognition applications in ADAS systems. Further, we will go through some more examples.

To begin with, there is a possibility to employ neural networks in driver identification. A recent research [A. Girma et al, 2019] demonstrate that Long-Short-Term-Memory (LSTM) model successfully predicts the identity of the driver based on the individual’s unique driving patterns learned from the vehicle telematics data. The proposed model is developed based on freely available vehicle telematics data collected from the OBD-II interface of vehicles.

Another research, performed by [T. Pamuła, 2012] presents the Traffic Flow Analysis Based on Real Data using Neural Networks. This work includes analysis of traffic data for determining classes of time series of traffic flow intensity for use in traffic forecasting employing neural networks. Such an approach is seen to be very promising in Intelligent Transportation Systems (ITS) applications.

A representative example of using convolutional neural networks in Classification of Telematics Car Driving Data is given by [G. Gao and M.V. Wüthrich, 2018]. Authors studied individual trips of different car drivers by considering their time series structures of speeds, acceleration, braking, and changes of angles. It was mentioned that by means of convolutional neural networks such tasks being performed successfully.

Going further, the bottleneck neural network approach is used by [G. Gao et al, 2019] in Claims Frequency Modeling Using Telematics Car Driving Data. The authors investigated claims frequency models, the low-speed interval, and the longitudinal acceleration rate. The logic underlying is that most accidents occur at low speeds.

Those are just e few of the large cohort of possible ideas and approaches in using neural networks based approaches to address various telematics challenges. It is pretty clear and obvious that such fruitful approaches are already intensively diffusing into telematics and would continue to do so in the future.

References:

  • T. Pamuła, Traffic Flow Analysis Based on the Real Data Using Neural Networks, 2012.
  • A. Girma et al, Driver Identification Based on Vehicle Telematics Data using LSTM-Recurrent Neural Network, 2019.
  • I. Livshin, Artificial Neural Networks with Java, 2019.
  • N. Purkait, Hands-On Neural networks with Keras, 2019.
  • G. Gao and M.V. Wüthrich, Convolutional Neural Network Classification of Telematics Car Driving Data, 2018.
  • G. Gao et al, Claims Frequency Modeling Using Telematics Car Driving Data, 2019.
  • https://www.navixy.com/blog/some-basic-principles-behind-neural-networks