Anomaly Detection in Salesforce’s Transactional Data Using Machine Learning Techniques

Abstract

This paper addresses the challenges posed by the volume and complexity of current healthcare systems, which necessitate the use of specific techniques for ensuring data security. Healthcare applications for Salesforce and Anomaly detection using Isolation Forest algorithm are the area of interest of this research work. Anomalous pattern detection is particularly important when it comes to differentiating an unusual pattern of behavior that can pose threats to security, for instance fraud or system failure. In this case, we use Isolation Forest algorithm in the Online Retail Dataset; a simulation of a real healthcare dataset. The fact that data preprocessing and feature engineering steps are tractable and do not require a labeled training data set along with its capability of accurately identifying anomalous data points in salesforce systems make it ideally suited for outlier detection in such systems. The method includes data preprocessing steps such as handling missing values and normalizing features, as well as new feature engineering to better recognize important customer patterns. The Isolation Forest model is then applied to identify the anomaly in the transaction data, achieving an accuracy of 93%, precision of 0.92, recall of 0.89, F1-score of 0.90, and AUC of 0.95. In line with our proposition, findings disclosed that Isolation Forest produced remarkably high accuracy and evaluation measures specifically in the area of outlier detection. Moreover, the model is used for an ongoing surveillance system for continual examination and learning to achieve a higher level of anomaly and outlier detection for the security of the healthcare systems. The research utilizes a simulated Salesforce healthcare dataset that is publicly available in order to remain compliant with data privacy regulations. An unsupervised Isolation Forest algorithm is used for the autonomous detection of anomalies without requiring pre-labeled data. The primary objective of this study is to develop and evaluate an unsupervised anomaly detection framework specifically tailored for Salesforce healthcare CRM systems. The novelty lies in combining context-rich feature engineering with the Isolation Forest algorithm to handle unlabeled and heterogeneous healthcare data. This framework offers a replicable methodology for enhancing fraud detection and operational security within healthcare CRM environments.