Robustness and reliability of a 1D-ConvNet in trajectory prediction with data augmentation from capacitive sensors

Nowadays, Machine Learning is an expanding branch of artificial intelligence under computer science. Machine Learning consists in a system, called Neural Network (NN), which can learn something from given data. For this reason, Machine Learning has a wide range of uses, for example from agriculture to spacecrafts. In this thesis work it will be used to predict a trajectory of a person with the knowledge of data from capacitive sensors, which were used in past as proximity sensors and now they are flourishing on the Internet of Things technology, because they are cheaper than infrared sensors used to detect and track movements. The goal of the thesis is to improve the robustness and reliability of a Neural Network in trajectory prediction, by adding noise to capacitive sensors data in Neural Network pre-process. This noise addition is a form of data augmentation, which leads to have more data, on which the Neural Network can learn. During data augmentation, it needs to be considered that the Neural Network can learn by heart, if there are too many data. This leads to the phenomena where the artificial mind learns better the data sequence than the ability to “reason” over the data. The 1Dimensional-Convolutional Neural Network (1D-CNN) is considered for experiments. The 1D-CNNs are simple and compact due to the fact they perform only 1D convolutions (scalar multiplications and addition). 1D CNNs have recently been proposed and immediately achieved the state-of-the-art performance levels in several applications such as personalized biomedical data classification and early diagnosis, structural health monitoring, anomaly detection and identification in power electronics and electrical motor fault detection. Using this setup, I got improved trajectories prediction for testing sets which are some similar and some different to the training and validation set. This result leads to an alternative method to filter the data, which feeds the NN.