AI-Powered Customer Churn Prediction in Banking with R-based Gradient Boosting Models and Power BI Dashboards R. Suyam Praba, P. Suganthi, J. Harrish Samraj, V. K. Arthi, M. Bhuvaneswari 2025 IEEE 1st International Conference on Innovations in Engineering and Next Generation Technologies for Sustainability Icinvents 2025, 2025 In the 21st century, Customer churn is a significant challenge in the banking sector, directly impacting revenue, operational efficiency, and long-term customer relationships. With increasing digital banking adoption and competition from fintech firms, banks must transition from reactive to proactive customer retention strategies. This study leverages machine learning techniques, specifically gradient boosting models such as XGBoost and LightGBM, to predict customer churn based on key demographic, behavioral, and financial attributes. The dataset undergoes rigorous preprocessing and feature engineering in <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{R}$</tex> to ensure data quality, eliminate inconsistencies, and optimize feature selection. Additionally, Power BI is integrated to create interactive dashboards that provide dynamic and actionable insights for financial institutions. The LightGBM model achieved an accuracy of 90.04 %, precision of 86.2 %, recall of 61.1 %, and F1-score of 71.4 %, outperforming the XGBoost model (accuracy 89%). By implementing predictive analytics, banks can transition from reactive to proactive churn management, optimizing customer retention strategies and enhancing business sustainability. This research highlights how machine learning-powered predictive modeling can provide a data-driven framework for improving banking sector customer loyalty and mitigating churn risks. The seamless integration of R-driven predictive analytics and Power BI visualization enables banking institutions to convert analytical findings into data-driven strategies, bridging the gap between quantitative modeling and executive decision-making [1].
AI-Driven Credit Decisioning for Robust Model Selection and Quantitative Performance Benchmarking P. Suganthi, R. Suyam Praba, D. Mythili, V. Smitha, K. Arnold 2025 IEEE 1st International Conference on Innovations in Engineering and Next Generation Technologies for Sustainability Icinvents 2025, 2025 In the changing world of financial services, AIbased credit decisioning has emerged as a key tool for making loan approval processes more efficient and accurate. Manual assessment techniques are typically time-consuming and inconsistent, resulting in operational inefficiencies. This research addresses strong model selection for AI-based credit decisioning and quantitative performance benchmarking to identify the best machine learning method for predicting loan approval results. The study entails preprocessing a historical loan application dataset and applying several machine learning models, such as Logistic Regression, Support Vector Machine (SVM), Decision Tree, Random Forest, and Gradient Boosting. Feature engineering, missing value handling, and data balancing methods are employed to enhance model performance. Hyperparameter tuning is done using RandomizedSearchCV to provide fine-tuned outcomes. To create a systematic benchmarking process, models are tested according to principal performance indicators including Accuracy, Precision, Recall, F1-Score. Comparability ensures that the most appropriate model for loan approval prediction is chosen to support data-driven decision-making. The research also determines principal features that affect loan approval and sheds light on the decision-making process. Through the inclusion of AI-powered models, the study shows an enhancement in the speed, uniformity, and accuracy of decisionmaking compared to conventional approaches. The results present a pragmatic template for financial institutions to upgrade loan processing capacity, minimize human error, and facilitate approval processes. The research sheds light on the significance of good model choice and quantitative benchmarking in credit decisioning process innovations and the promotion of efficient loan approvals in finance.
An Application of NB-GA Model: A Study of Logistics Performance and Economic Attributes Bhavanam Amarnath Reddy, Anjan Kumar Reddy Ayyadapu, Shivagond Nagappa Teli, R Suyam Praba, V. O. Kavitha, Roop Raj International Conference on Intelligent Algorithms for Computational Intelligence Systems Iacis 2024, 2024 These days, the majority of logistics systems depend on AI strategies to enhance services and increase outcomes. Online marketplaces are an exception to this rule. The diversity and complexity of resource allocation and job scheduling, however, remain formidable challenges in dynamic systems. Therefore, solid algorithms supporting complex models are required. Adding sophisticated models to scheduling systems is one way to boost logistical efficiency. Therefore, controlling system resources is still crucial for improving job scheduling while considering the interaction implications and logistics system needs. Model training, feature selection, and preprocessing are all dependent on sequencing. Normalization is a part of the preprocessing phase. According to the Normalized Z-Scores and the Gaussian Distribution. PCA and correlation are the two main components of feature selection. One way statisticians quantify the degree of linear relationship between two variables in a set is with the Pearson correlation coefficient. When dealing with data that is multidimensional, principal component analysis (PCA) is one of the most popular multivariate statistical methods. Training an NB-GA requires exact regulation of attributes. Modern NB and GA algorithms are quaint when compared to this technique. The findings showed a significant improvement in accuracy, with a grade of 97.22%.
Exploring Financial Literacy's Impact on Preventing Economic Crimes: A Random Forest Analysis Naveen Pol, Anand Guled, T. Manikumar, R Suyam Praba, E. K. Arulkarthick, Roop Raj Proceedings of International Conference on Sustainable Computing and Integrated Communication in Changing Landscape of AI Icscai 2024, 2024 Financial crime poses a substantial menace to individuals, businesses, and economies on a global scale. In spite of attempts to hinder such unlawful acts, the intricate characteristics of deceitful activities require inventive methods for identification and reduction. This study aims to investigate the correlation between financial literacy and the prevention of economic crime by utilising sophisticated machine learning techniques, notably Random Forest Analysis. The approach we present combines data collection, preprocessing, feature selection, model construction, and Random Forest Analysis to forecast economic crime using financial literacy levels. In contrast to previous studies, our methodology provides several benefits, such as thorough feature selection, resilient model training, and exceptional predicted accuracy. The assessment of the suggested system reveals outstanding outcomes, with an accuracy of 0.94, precision of 0.92, recall of 0.90, F1-score of 0.95, and AUC of 0.92. These findings emphasise the efficacy of utilising financial literacy to reduce the dangers of economic crime and showcase the promise of modern machine learning approaches in tackling intricate social issues.
