Stochastic Gradient Boosting for Robust Twitter Spam Detection: An Ensemble Learning Approach
Sanjay Kumar Suman, Pooja Sahu, Monika Khatkar, K Haribabu, S. Poongodi, Homera Durani
Abstract
Here, we concentrate on one specific use case: Twitter identifying spam using the Stochastic Gradient Boosting (SGB) technique. In order to improve the predictability of prediction models, combination methods of learning are investigated for merging several ineffective learners into a single robust system. To be more precise, improving is a process that results in better models for predicting by building upon the mistakes of earlier weak learners. One kind of boosting called Gradient Boosting is applicable to both classification and regression problems since it continuously enhances effectiveness by reducing the loss function. To improve generalization and decrease correlation among weak learners, Stochastic Gradient Boosting (SGB) subsamples the training dataset to inject unpredictability. Among SGB's many benefits discussed in this chapter are its adaptability to non-linear connections, its scalability to deal with massive datasets, and its resilience in the face of inaccurate or missing data. It proposes a method that uses SGB to identify Twitter spam by making use of lightweight characteristics. When compared to more conventional machine learning models, SGB achieves better accuracy, precision, and recall. These models include Logistic Regression, Random Forest, and Support Vector Machine. Because of its repeated and random sample processes, SGB is able to achieve far higher detection rates than competing approaches.