Deep Learning Techniques: A Case Study on Comparative analysis of Various Optimizers to Detect Bots from CRESCI-2017 Dataset
Abstract
Deep learning is the most popular and stimulating field of machine learning that provides tremendous signs of progress on various applications such as speech recognition, natural language processing, computer vision classification, medical image analysis, detecting financial frauds, social network spam filtering and so on. It has turned out the most exciting research topic in recent years since it produced promising responses for a large scale of data. However, the lack of a better understanding of its behavior, architectural models, optimization techniques, activation functions, frameworks still keeps the field more challenging, and there can be a great extent of improvements. This article aims to elaborate on deep learning theory, various deep learning architectures, optimization, activation functions. A deep learning keras model is built using python scikit learn and experimented with various optimizers such as ADAM, SGD, RMSprop, Adadelta, Adamax, Adagrad, Nadam to detect bot accounts from CRESCI-2017 twitter dataset issued by Indiana University. RMSProp shows 98.90% of accuracy with 0.12% of loss in contrast with other optimizers. The comprehensive survey of this article will give a complete insight into the deep learning challenges and renders how these challenges proffer support to transform into a productive future of research study.