Predicting Consumer Behaviour in Digital Marketing Using Scalable Machine Learning Models

The rapid growth of e-commerce platforms has spawned large, non-homogenous and highly volatile volumes of consumer data, thus generating a pressing need to predictive models able to condense meaningful behavioural evidence. Traditional statistical models are often inadequate to capture the nonlinear relationships between browsing behaviour, discount sensitivity, opinion on reviews and demographic characteristics that characterize consumer behaviour in the era of modern society. This study outlines a comparative framework that uses scalable machine-learning algorithms like Support Vector Machine (SVM), Categorical Boosting (CatBoost), Extreme gradient boosting (XGBoost) & Backpropagation Artificial Networks (BPANN) to predict purchase decisions based on the Flipkart Consumer Behaviour dataset which includes more than 100,000 user sessions. The preprocessing pipeline that we designed includes z-score normalisation, Synthetic Minority Oversampling Technique (SMOTE) and Isolation Forest to address the issue of class imbalance and reduce the impact of data noise. The models were optimised on a ten-fold cross-validation structure, hyperparameter optimisation done through Grid & Random search and performance was measured on Accuracy, Precision, F1-score, Recall & ROC-AUC. The findings show that CatBoost provides the best predictive performance with a ROC-AUC of 0.982 and a recall of 94.1 which outperforms alternative learners in the ability to deal with categorical variables, but XGBoost has a high generalisation potential. SVM model resulted in superior precision (94.8%), and the BPANN retained nonlinear dependencies but did not rival in scalability. These results emphasise the effectiveness of boosting-based ensembles in consumer behaviour modelling and show that they are useful in precision-based digital marketing approaches.

Перевод пока недоступен