Intelligence Platform | TimothyMaina

#	Module	Model Architecture	Key Metric	Last Run	GitHub	Status
01	Sales Intelligence	ARIMA / Prophet	R² 0.84	2 min ago	→ Repo	LIVE
02	Churn & Segmentation	Logistic Reg · K-Means	Recall 0.88	3 hr ago	→ Repo	LIVE
03	Fraud Command Centre	Random Forest · SMOTE	F1 0.91	2 min ago	→ Repo	ALERT
04	Marketing ROI	Markov Chain · RF	R² 0.79	1 hr ago	→ Repo	LIVE
05	Supply Chain	Prophet · EOQ · SS · ROP	MAPE 8.2%	30 min ago	→ Repo	IN PROG
06	Credit Risk	XGBoost · SMOTE	F1 0.76	1 hr ago	→ Repo	IN PROG
07	Logistics Optimization	Random Forest Regressor	R² 0.70	15 min ago	→ Repo	IN PROG

Sales Intelligence — Superstore

4-year Superstore dataset · ARIMA / Prophet time-series forecasting · Tableau + Excel

ARIMAProphet Completed12-Month Forecast

Prediction Inputs

Region

Explainability — Feature Importance

Run a forecast to see feature importance

Revenue Trend + 12-Month Forecast

What-If Scenario Simulation

Discount applied (%)10%

Marketing spend increase (%)+20%

Headcount growth (%)+5%

⟳ Scenario impact: Est. +$84K revenue with current settings

Loss-Making Sub-Categories — Flagged

Sub-Category	Region	Margin	Signal
Tables	East	-8.4%	Unprofitable
Bookcases	Central	-2.1%	Unprofitable
Supplies	West	-0.9%	Marginal
Machines	South	1.2%	Watch

Insight: Tables sub-category is consistently unprofitable across East region. Recommend pricing review or product discontinuation.

Customer Churn & Segmentation

E-commerce dataset · Logistic Regression (high-recall) · K-Means RFM Clustering · Tableau

Logistic RegressionK-Means RFMCompleted · Recall 0.88

Customer Profile Inputs

Days since last purchase 45d

Total orders placed 3

Total spend ($) $120

Refund rate (%) 10%

Churn Probability

—

RFM Segment Distribution

Model Feature Importance

Days inactive38%

Refund behaviour27%

Order frequency19%

Total spend (monetary)16%

Model insight: High inactivity combined with elevated refund rate is the strongest predictor of churn. Customers inactive 90+ days with >20% refund rate are 4× more likely to churn.

Retention Action Playbook

Segment	Risk	Action
Champions	Low	Loyalty rewards + referral
Loyal Customers	Low	Upsell + NPS survey
At-Risk	High	Win-back email + 15% off
Inactive	Very High	Reactivation campaign
New Customers	New	Onboarding sequence

Fraud Detection Command Centre

Random Forest · SMOTE class balancing · Behavioural feature engineering · Tableau

Random ForestSMOTE ALERT ACTIVEF1 0.91 · Completed

Transaction Inputs

Transaction amount ($) $250

Transactions per hour 2

Distinct devices used 1

IP repetition score 1

Fraud Probability

—

Fraud Trend — Last 30 Days

Feature Importance — Fraud Signals

Transaction velocity (burst pattern)41%

Device count (account takeover)26%

IP repetition (repeat offender)21%

Transaction amount anomaly12%

Model insight: High-frequency burst patterns are the primary fraud signal. Velocity spikes of 15+ transactions/hour flag with 94% precision in the Random Forest model.

High-Risk Entities — Live Flag

Entity	Score	Primary Signal	Action
Card_7821	0.94	Velocity burst	Block
IP_192.11	0.87	IP repeat	Block
User_4420	0.73	Multi-device	Flag
Card_3312	0.61	High amount	Monitor
User_8891	0.28	Low risk	Pass

Marketing ROI Optimization

Markov Chain Attribution · Random Forest revenue prediction · Power BI · R² 0.79

Markov ChainRandom Forest Completed · R² 0.79

Campaign Inputs

Total campaign spend ($) $5,000

Cost per acquisition ($) $45

Conversion rate (%) 4%

Channel touchpoints 3

Predicted Revenue

—

Markov Chain Attribution — Channel Contribution

Budget Reallocation Recommendation

Model insight: CPA is the dominant driver of revenue (feature importance: 38%). Channels with lower CPA yield disproportionately higher returns under the Markov attribution framework.

Channel	Current Spend	Attribution %	CPA	Predicted ROI	Recommendation
Email	$1,200	34%	$12	8.4×	↑ Increase budget
Paid Search	$2,100	28%	$38	3.1×	→ Maintain
Social	$900	22%	$55	2.4×	→ Maintain
Display	$800	16%	$89	1.2×	↓ Reduce budget

Supply Chain & Inventory Optimization

Prophet · ARIMA · Safety Stock · Reorder Point · EOQ · 1,900+ stockout events identified

ProphetARIMA EOQ · SS · ROPIn Progress

SKU Inputs

Avg daily demand (units) 50

Lead time (days) 7d

Demand std dev σ 10

Service level target 95%

Holding cost ($/unit/yr) $2

Order cost ($) $50

—

Safety Stock (units)

—

Reorder Point (units)

—

EOQ (units/order)

—

30-Day Forecast (units)

Demand Forecast — Prophet + Historical

Customer Credit Risk Assessment

XGBoost · SMOTE · Composite risk scoring · 18.3% default rate vs 9.5% benchmark

XGBoostSMOTE Feature EngineeringIn Progress · F1 0.76

Borrower Profile Inputs

Annual income ($) $45,000

Debt-to-income ratio (%) 35%

Credit score 680

Loan amount ($) $15,000

Default Probability

—

Risk Tier Distribution

Feature Importance — XGBoost

Debt-to-income ratio31%

Credit score band28%

Loan-to-income ratio22%

Income level19%

Logistics & Delivery Optimization

Random Forest Regressor · R² 0.70 · Traffic & weather proxy features · Driver performance analysis

Random ForestFeature Engineering In Progress · R² 0.70

Shipment Inputs

Route distance (km) 250km

Historical delay rate (%) 15%

Weather risk score (1-10) 3

Traffic congestion (1-5) 2

Carrier

Delay Probability

—

Delay Distribution — Historical Buckets

Feature Importance — R² 0.70 Model

Driver performance score33%

Route conditions27%

Weather risk proxy22%

Traffic congestion proxy18%

Key finding: Data leakage features were identified and removed (inflated R² to ~0.99). Final R² 0.70 represents a realistic, production-aligned baseline.

Executive Dashboard

Live System Alerts

Revenue trend + 12-month forecast

Churn distribution by RFM segment

Fraud incidents — 13-week trend

Inventory stockout risk by warehouse

Module Performance Registry