A Machine Learning Method for Marketing Prediction of Consumer Behaviour
Аннотация
In the world of digital marketing, accurately predicting consumer behaviour is crucial to creating focused and effective tactics. This paper presents a method that uses machine learning to forecast consumer behaviour based on a large dataset consisting of 50,000 records of consumers. Demographic data, purchase history, interactions with online and mobile apps and indications of social media participation are all part of the collection. A strong data preparation pipeline was used which included filling in missing values using imputation, normalizing numerical data using z-scores, encoding categorical features using one-hot encoding and reducing dimensionality through Principal Component Analysis (PCA). The principal component analysis (PCA) greatly improved computing efficiency while keeping crucial information; it kept over 91% of the overall variance with only 12 principal components. The classification methods like Logistical Regression, Random Forest and XGBoost were tested using 10-fold cross-validation. F1-score, AUC-ROC, accuracy, precision and recall measured model performance. XGBoost performed best with an AUC-ROC score of 0.95 suggesting strong purchaser-non-purchaser discrimination. The best method made real-time predictions via a scalable API. Data distribution changes were identified after deployment by weekly Kullback-Leibler (KL) divergence monitoring. Model drift shown by a rise in Week 6 necessitated retraining. The results of this study demonstrate that machine learning is capable of marketing forecasting and offers an adaptable, scalable framework for involving consumers and making strategic decisions.