Pump and dump delection on crypto currencies using computer vision

Abstract

Inspired by the immense success shown by artificial neural networks in computer vision on images classification, we propose a novel framework to detect one of the rife fraudulent financial manipulations in crypto currency trading world known as pump and dump. The representation of crypto currency financial charts was re-imagined ameliorating the classification by taking advantage of some of the very recent advancements of time series to spatial encoding techniques of Gramian Angular Field (GAF), Markov Transition Field (MTF) and Recurrence plots (RP) that are capable of spatially encoding the temporal financial time series data in the form of images. Encoded images were then used to train several convolutional neural network architectures which have been able to achieve a very high precision, recall and F1 values close to 99% over the unseen data for the above classification task. This is one of the first of such researches in pump and dump detection in crypto currencies using computer vision. This approach has the potential to be extended in detecting predefined shapes of time series charts.