Filtering spam e-mails with a context-independent approach based on angle transformation

In this study, an angle-based approach developed for the detection of unwanted emails defined as spam has been examined. The text contents of the emails were transformed into Unicodes and treated as one-dimensional signals. Angle information between the values on this signal has been calculated. The obtained angle signal has been used as a histogram feature vector specific to each email. Various machine learning methods, including Naive Bayes (NB), Support Vector Machines (SVM), K-Nearest Neighbors (Knn), and Random Forest (RF), were used to test the success of this approach. These classification processes were conducted using the open-source Weka program and evaluated through 10-fold cross-validation. The results indicate that the Knn method achieved a success rate of 94.2%. Other methods also showed acceptable success rates. Furthermore, the use of different values for parameters such as uL and uR emphasized the flexibility of the angle approach in obtaining different patterns. Particularly, high success was achieved with parameter values uR=1 and uL=1. However, it was emphasized that these parameters might require different evaluations in various datasets. This study demonstrates that an angle-based approach using character Unicode values is an effective way for spam detection. This approach offers an alternative method to traditional text analysis in the fight against spam.