Spam Detection Using Genetic Algorithm Optimized LSTM Model

Abstract

The advancement in technology over the years has resulted in the increased usage of SMS which in turn has provided certain groups a chance to exploit this service by spreading spam messages to consumers making it difficult for people to receive important information and also possessing a threat to their privacy. There are numerous machine learning and deep learning techniques that have been used for spam detection and have proved to be effective. But in deep learning techniques, it is essential to fine-tune the hyperparameters which requires excessive computational power and time, making the process less feasible. The proposed work aims at reducing this computational barrier and time by using Genetic Algorithm in order to select the key hyperparameters. A randomly generated population of LSTM models was created and further generations were produced following the different stages of the genetic algorithm multiple times until the terminal condition was met, and the performance of each candidate solution was evaluated using a chosen fitness function. The most optimal configuration was obtained from the final generation which is used to classify the messages. Four metrics, namely the accuracy, precision, recall and f1-score were used to analyze the model’s performance. The experimental results demonstrate that the Genetic Algorithm optimized LSTM model was able to outperform the other machine learning models.