Dashboard

Live dashboards

Four public Streamlit apps from completed data-science work. Sales, churn, fraud detection, and stock analysis sit in that order below.

Sales Analysis Dashboard Customer Churn Dashboard Fraud Detection Dashboard Stock Analysis Dashboard

Sales Analysis Dashboard

Filter the book, inspect distributions, score net-sales regressors, and compare monthly forecasts — then open the live app to use it.

Open live dashboard View source

https://mwperrinejr-completed--online-retailer-sales-analysisapp-092rwf.streamlit.app Open live app

Sales Analysis Dashboard

European product orders, 2019–2020 — EDA, net-sales regressors, and monthly forecasts.

Orders

1,000

Total net sales

€18,992,428

Avg net sales / order

€18,992

Blended margin

16.8%

Overview & EDARegression ModelsTime Series Forecast

Distribution of sales

Sales (EUR)Frequency

Distribution of COGS

COGS (EUR)Frequency

Sales vs COGS by country

Monthly net sales: forecast vs actual

Train Actual ARIMA

Open live dashboard

Static preview of the deployed app. Click the frame or open the live dashboard.

Open live dashboard

What’s inside

Overview & EDA

Filter 1,000 orders by country, category, manager, device, and date. Histograms, sales vs. COGS by country, and period-over-period change.

Regression Models

Predict net sales from categorical attributes only — Linear, Huber, and XGBoost — with actual-vs-predicted plots and holdout metrics.

Time Series Forecast

Monthly net-sales forecasts from irregular order dates, comparing a manual ARIMA spec with auto_arima on a holdout window.

Customer Churn Dashboard

Explore 10,000 bank customers, review a persisted ANN on held-out test performance, and score one customer at a time.

Open live dashboard View source

https://completed-projects-i5vraxfwegymcjdyewtywt.streamlit.app Open live app

Customer Churn Dashboard

Bank customer churn — EDA, an ANN classifier, and a single-customer predictor.

Customers

10,000

Churn rate

20.4%

Avg balance

€76,486

Avg credit score

651

Overview & EDAModel PerformancePredict a Customer

Distribution of credit scores

Credit scoreFrequency

Salary by gender

Average balance: stayed vs churned

Churn rate by geography

Open live dashboard

Static preview of the deployed app. Click the frame or open the live dashboard.

Open live dashboard

What’s inside

Overview & EDA

Filter by geography, gender, age, and active status. Credit-score distributions, salary by gender, balance by churn, and churn rate by country.

Model Performance

A 6-6-1 ANN trained once on the full set. Test accuracy 85.8%, ROC AUC 0.864, with a live threshold slider on the confusion matrix.

Predict a Customer

Interactive form that mirrors the notebook’s manual prediction cell — credit score, geography, products, and a churn probability.

Fraud Detection Dashboard

Explore 6.36 million mobile-money transactions, review a balanced logistic regression on a held-out test set, and score a single transfer.

Open live dashboard View source

https://mwperrinejr-completed-projects-frauddetectionapp-raaknk.streamlit.app Open live app

Fraud Detection Dashboard

Mobile-money transaction fraud — EDA on 6.36M transactions, a logistic regression classifier, and a transaction scorer.

Total transactions

6,362,620

Fraud rate

0.129%

Fraudulent transactions

8,213

Flagged by the source system

Overview & EDAModel PerformancePredict a Transaction

Transaction volume by type

Fraud rate by transaction type

Log(amount + 1)

log(amount + 1)Frequency

Fraud count per time step

Open live dashboard

Static preview of the deployed app. Click the frame or open the live dashboard.

Open live dashboard

What’s inside

Overview & EDA

Exact aggregates over 6.36M rows: volume and fraud rate by type, amount distributions, frauds over time, and a zero-balance-after-transfer signature.

Model Performance

Logistic regression with class_weight='balanced'. Test recall 93.6% and ROC AUC 0.989 — accuracy is 94.7%, but precision/recall are the metrics that matter at 0.13% fraud.

Predict a Transaction

Score a single mobile-money transfer with the same fields as the standalone form, plus a live decision threshold.

Stock Analysis Dashboard

Compare prices, SMA crosses, Bollinger bands, and MACD across five names, then rebuild an inverse-volatility book and stress it with Monte Carlo.

Open live dashboard View source

https://completed-projects-d4kxlppp8zwtqry6lpaa8h.streamlit.app Open live app

Stock Analysis Dashboard

Technicals and a five-name tech book for AAPL, MSFT, GOOGL, NVDA, and TSLA. Every number is in-sample — it describes this window, not a forecast.

Universe

5 names

Window start

Jan 2020

Random books

25,000

Bootstrap paths

5,000

PriceMoving averagesBollinger bandsMACDWeightsMonte Carlo

Growth of $1 at the start of the window

AAPL MSFT GOOGL NVDA TSLA

10-day vs 50-day SMA

Bollinger bands (20-day, 2σ)

Feasible set (25,000 long-only books)

Open live dashboard

Static preview of the deployed app. Click the frame or open the live dashboard.

Open live dashboard

What’s inside

Price & technicals

Adjusted close vs growth of $1, 10/50-day SMA golden and death crosses, 20-day Bollinger bands, and MACD (12, 26, 9) plotted side by side.

Portfolio weights

Equal-weight, inverse-volatility, and inverse-variance books. Capital weights and risk contribution are shown separately so NVDA and TSLA do not hide in dollar shares.

Monte Carlo

25,000 long-only random books for the feasible set, plus 5,000 one-year bootstrap paths on the inverse-volatility rule from a $10,000 start.