Hi, I’m Olivia Justine Nanteza; I turn data into solutions.

Project Context

This project explored Shake It Co. Ltd.’s retail performance using a structured dataset of 30,000 transactions across customers, products, payment methods and delivery modes, with the aim of assessing financial health, behavior, risks and opportunities.

Customers & Demographics

Gender/age were often Not Applicable (≈66%), masking some segmentation depth. Among captured values, male, female and non-binary customers were fairly balanced, hinting at broad appeal and the upside of improving data capture for personalization.

Payment & Product Preferences

Mobile Money dominated payments (13,678), ahead of Cash (8,899), Credit (4,458) and Debit (2,965), clear signal to lean into digital payment incentives. Product-wise, Electronics, Apparel, and Home Goods each contributed ~10k transactions, giving a well-balanced portfolio.

Financial Health

The business generated $15.79M revenue, $12.34M cost, and $3.45M profit, yielding a healthy ~21.85% margin. However, some transactions posted heavy losses (e.g., −$745), meriting root-cause analysis on pricing, returns, and logistics.

Resilience via Diversification

Revenue and profit were evenly distributed across regions, customer segments, categories, branches and delivery modes, reducing over-reliance on any single dimension and hardening the model against shocks.

Seasonality & Trends

A predictable September–October dip appears in both revenue and profit; otherwise performance is stable year-over-year with modest growth, classic signs of a mature, steady operation ready for targeted optimization.

Recommendations

• Push digital payments: mobile-money perks/loyalty to cement the channel.
• Fix negative-profit outliers: investigate SKUs, discounts, and return flows.
• Seasonal plays: promo bundles or events to counter the Sep–Oct slump.
• Better data capture: collect age/gender to unlock precise targeting.

Project Overview

This team project addressed late diabetes diagnosis, where patients are often identified only after serious symptoms and complications emerge. We built a predictive tool that leverages routine patient checkup data to flag high-risk individuals earlier and enable timely intervention.

Background & Problem

Diabetes is a fast-growing global health challenge and often goes undiagnosed until complications appear. The missed opportunity is to use already-collected routine data to predict risk before symptoms develop, improving outcomes and reducing costs through early action.

About the Data

We analyzed 1,000 anonymized patient records (844 diabetic, 103 non-diabetic, 53 pre-diabetic) with 12 health factors including HbA1c, cholesterol, BMI, age, and others, sourced from a medical records dataset on Kaggle.

Predictive Modeling

Using Python (Pandas, scikit-learn), we pre-processed the data and evaluated several models. A Random Forest classifier achieved 99.6% accuracy, predicting diabetic, pre-diabetic, and non-diabetic classes.

Key Insight , HbA1c

HbA1c emerged as the strongest predictor of diabetes risk, more informative than age or BMI, and it’s already part of routine labs, making implementation practical with no extra testing burden.

Implications & Recommendations

The model enables proactive identification of high-risk patients and supports early intervention programs. We recommend integrating the tool into EMR systems for automated flagging during visits and piloting structured lifestyle support for the pre-diabetic cohort.

Limitations & Next Steps

Limitations include few pre-diabetic cases, little lifestyle/family history data, and no longitudinal tracking. Next steps: enrich data (diet, activity, family history, wearables), expand population coverage, and validate over time series for clinical deployment confidence.