Character-Prediction-Using-Mobile-Sensors
Deep Learning has impacted diverse fields, such as healthcare, fraud detection, natural language processing (NLP), and others. The contributors of this repository wanted to create an impression and a mark in this field.
We tried our hands out on Character Prediction using Smartphone Sensors. The increased usability and database of smartphone users gave us the idea of creating letters A-Z and digits 0-9 in three-dimensional space using smartphone sensors like gyroscope, accelerometer, magnetometer and linear accelerometer. These sensors provide us with time-series data. Utilising this data and the boundless possibilities of the neural network on character recognition, we built and analysed various models and settled at the Bidirectional LSTM model that provides a training accuracy of 93.60% and testing accuracy of 89.51%.
Since no state-of-the-art dataset was present for the problem statement, we wanted to tackle the paper's authors (contributors to the repository) made the dataset from scratch. The dataset is made publicly available to fellow innovators and coders who share the same curiosity towards solving a new problem with a different approach. One can access the data here: Char-sense Dataset
Our work has been published in the International Conference on Applied Scientific Computational Intelligence using Data Science (ASCI 2020). The link to which is: Character Recognition on Time Series Data collected from Smartphone Sensors