Comparative Study of Machine Learning Algorithms for Predictive Analytics in Web Applications

This paper delves into the internal workings of some machine learning algorithms that could be implemented to online predictive analytics, or to e-commerce predicting user behaviour. The first one uses a similar experimental design to estimate and quantify different models with respect to the run time and predictive power. Data on user interactions, purchases, and demographics are part of the 75,000 entries and 28 variables that make up the semi-public collection. Normalisation, SMOTE encoded class balancing, and encoding are the preprocessing strategies to enhance the quality of the data. Several methods are employed in the study, such as Decision Tree, Logistic Regression, Support Vector Machine, K-Nearest Neighbours, Naive Bayes and Artificial Neural Network (ANN). Among them, ANN is the most effective as it has an accuracy of 92.6 percent, and ROC-AUC 0.96. The findings indicate that more complex models are unable to model non-linear effects at online data. It is critical to pick models in accordance with deployment needs, and there is a trade-off between accuracy and computing cost. This study's findings provide credence to the idea that, when set up correctly, classical machine learning algorithms can hold their own against state-of-the-art AutoML. What's more, these algorithms give us a scalable predictive model that works great for real-time e-commerce.