3 results across projects and posts.
An end-to-end ML product for identifying at-risk telecom customers before they leave, from raw data exploration through to a deployed Streamlit prediction app. Built on 594K synthetic training rows from the IBM Telco dataset. The key finding: cost-sensitive modelling delivers 25x more business value than a standard classifier at identical AUC.
A predictive scoring system built to proactively identify at-risk key accounts during COVID-19, enabling account managers to intervene before contracts were lost and achieving 100% Key Account retention across the portfolio.
An unsupervised ML pipeline combining K-Means clustering with statistical outlier detection to identify non-compliant sales behaviour across a 300,000-outlet retail network, protecting incentive payouts and strengthening reporting integrity.
