Prediction of Clients Based on Google Analytics Income Using Support Vector Machines

This study aims to deploy Support Vector Machines (SVMs) to classify clients within the user base of the IMOLKO company website, predicated on the analysis of clickstream behavior. The study conducted several experiments using Monte Carlo cross-validation, encompassing diverse training and testing data proportions. Model performance was evaluated using parameters such as accuracy, sensitivity, specificity, positive predictive value, negative predictive value, and F1 score. The results indicate that the SVMs consistently performs well across multiple runs, as evidenced by the low standard deviations associated with the evaluation metrics. It suggests that the results are reliable and not strongly influenced by random variations. The findings indicate that SVMs is an acceptable classification technique for predicting client status in the context of IMOLKO C.A. However, it is worth noting that although the model effectively predicts non-customers, the possibility of false positives exists, which reduces the percentage of F1 scores. The imbalance in the database, with a significantly higher number of non-clients compared to clients, may be impacting the method's efficiency. A balanced database, where each class has a similar number of examples, is desirable in classification tasks to avoid biases towards a dominant client class and ensure accurate decisions for all client classes. In conclusion, SVMs show promise as a reliable classification technique for predicting client status in the IMOLKO C.A. context. However, addressing database imbalance and conducting further research are imperative to enhance the performance of the models.