Investigating the Effectiveness of Word2Vec for Spam Detection Using Lazy Predict Library

Abstract

The proliferation of Email for exchanging information and messages through internet coincides with a significant rise in unsolicited email (spam),making it increasingly difficult for users to manage their in-boxes and identify legitimate messages. A multitude of detection methodologies have been established and refined to address the deluge of unsolicited electronic mail messages. These approaches encompass knowledge-based techniques, clustering algorithms, learning-based models, heuristic algorithms, and potentially other methodologies. It is noteworthy that while numerous advancements have been made, none of these detection models or techniques have achieved perfect predictive accuracy. Within the domain of spam email detection, machine learning(ML) and deep learning(DL) algorithms have emerged as the most effectual methodologies amongst the plethora of models proposed. Choosing the optimal model for a ML problem can be a challenging task.To solve this problem, we start by converting the email text into vector features using word2Vec and applying various machine learning classifiers on the dataset using Lazy Predict classifiers with default parameters for ML models, We'll then evaluate our basic model's performance after fine-tuning Word2Vec hyperparameters. Here basic model means "Model without parameters", we chose the best models, then applied a hyper parameter adjustment to them. This investigation explores the efficacy of word2Vec with ML in spam email classification. The proposed approach achieved a commendable accuracy of 0.99, signifying its potential as a valuable tool for enhancing spam detection capabilities.