Download PDFOpen PDF in browser

Deployment of Deep Learning Models to Mobile Devices for Spam Classification

EasyChair Preprint 2412

6 pagesDate: January 18, 2020

Abstract

With the advent of deep learning, all the applications in the real world are taught to ingest deep learning to make them work better and fast. One such application is implemented in this work for detecting spam messages. A Binary Classification model which has negative(0) or positive(1) is trained on the deep learning python library Keras. The trained model is transformed to a graph using the graph transformations of tensorflow and is converted to a protobuf file to be implemented on android or iOS devices. This will call for a huge change in the way spam messages are dealt. Instead of looking into the spam messages in an algorithmic way just with keywords, it is dealt with experience of learning and predicting whether a text message is spam or not spam. The training was performed multiple times on resource-deficient device and hyper-parameter optimization was performed to enhance the training accuracy to 99.87%. Then sequential modeling, graph transformation and android development were performed. The mobile application uses the trained model to classify if a message is spam or not in real time (even without any internet access) with a test accuracy of 98.7%. Our simulation shows that a model with an embedding layer (size 128), an LSTM layer (size 64, dropout 0.2) and a dense layer (sigmoid activation) yields the highest classification performance. Also, we conducted comparative classification evaluation with state-of-the-art methods and our model has achieved higher accuracy.

Keyphrases: Android, deep learning, mobile device, spam classification, text analysis

BibTeX entry
BibTeX does not have the right entry for preprints. This is a hack for producing the correct reference:
@booklet{EasyChair:2412,
  author    = {Ameema Zainab and Dabeeruddin Syed and Dena Al-Thani},
  title     = {Deployment of Deep Learning Models to Mobile Devices for Spam Classification},
  howpublished = {EasyChair Preprint 2412},
  year      = {EasyChair, 2020}}
Download PDFOpen PDF in browser