Devam Shah

← All projects

Statistics

Data Science Projects Portfolio

Four end-to-end analyses covering statistics, supervised learning and unsupervised learning, each from raw data to business recommendation.

4 analyseshypothesis testing, classification, regression and clustering

View the code on GitHub ↗
Stack: Python, pandas, scikit-learn, statsmodels, SciPy

Live demo

A working version of this project will run right here.

Coming soon

Plan: One mini-demo per analysis: a test-results explorer, a renal risk view with an ROC threshold slider, advertising-budget regression sliders and a cluster viewer.

TODO: build and embed the demo

The four analyses

  • 01 US Customer InsightsHypothesis testing (t-test, ANOVA, Chi-Square, Pearson) on 10,675 retail customers. Spending is uniform across gender, education and state, but married customers with pets spend more.
  • 02 Renal Health Risk Prediction20,000 clinical records (13,847 after cleaning). Logistic Regression, KNN, Decision Tree and Random Forest with GridSearchCV and ROC-AUC threshold tuning. Best: Random Forest, ROC-AUC 0.8257. Top predictors: GFR and creatinine.
  • 03 Advertising Budget & SalesMultiple linear regression on 200 observations (ISLR). Best TV+Radio model: test R² = 0.89, RMSE ≈ 1,030 units. Newspaper spend is statistically insignificant.
  • 04 Student Social Network SegmentationK-Means and hierarchical clustering on 15,000 profiles (38 features), with PCA and silhouette scoring. 4 clusters; 528 students flagged in an At-Risk cluster.

Code and write-ups