A machine learning framework for customer value identification based on transaction behaviour

Abstract This research focuses on the critical task of identifying high-value customers by analysing the spending behaviours of customers of Sprocket Central Pty Ltd through the application of advanced machine learning techniques on the dataset collected dataset obtained from the KPMG Data Analytics Consulting Virtual Internship, accessed via Kaggle. This comprehensive dataset comprises detailed consumer transactional records, including monetary expenses, purchase frequency, recency of transactions, and demographic variables and so on. So, this study implements two widely used classification algorithms here to find their effectiveness which are Random Forest Classifier and Logistic Regression model . These models were selected to represent contrasting approaches in predictive analytics of the collected data, with Random Forest exemplifying ensemble-based along with non-linear modelling and Logistic Regression representing a traditional as well as interpretable linear method used by most of the analysts. The objective of this research was to predict which customers fall into the high spender category that is pivotal for maximizing profitability and optimizing marketing efforts by Sprocket Central Pty Ltd . Model performance was rigorously evaluated using a suite of standard classification metrics, including accuracy, precision, recall, F1-score, and the Area Under the Receiver Operating Characteristic Curve (ROC AUC) to understand the efficiency of model performance as well as the expected output from the dataset analysis. The Random Forest model consistently outperformed Logistic Regression across all metrics, achieving an AUC score of 0.87, signifying excellent discriminatory power between high and low spenders from the dataset. In contrast, Logistic Regression yielded a respectable but comparatively lower AUC of 0.81 which is reflecting limitations in capturing the complex, non-linear relationships inherent in consumer spending behaviour in constantly changing business environment. Here, the feature importance analysis within the Random Forest model revealed that monetary value of past transactions is the most influential predictor, followed by transaction frequency, customer age, and recency of purchase. This insight confirms that both behavioural and demographic factors play a crucial role in defining customer value for business to ensure their efforts to be put in the right place. Furthermore, the challenges that has been encountered throughout the research such as data imbalance, where high spenders constitute a minority of the population and addresses these through careful metric selection rather than oversampling techniques. This shows a future scope for further research aligning this topic. Overall, the research underscores the substantial benefits of employing ensemble machine learning approaches in customer segmentation tasks through offering similar businesses an enhanced predictive accuracy and more granular customer insights. It also supports the integration of such models into business intelligence and customer relationship management systems to facilitate targeted marketing campaigns, efficient resource allocation, and improved customer retention strategies which can be effectively done through similar dataset analysis. The findings contribute to the growing body of knowledge on machine learning applications in marketing analytics and provide a foundation for future research to refining model robustness, interpretability, and deployment in dynamic retail environments by businesses.