MBA Capstone · AI & Data Science · SRM Institute of Science and Technology

AI-Driven Demand
Forecasting for Sustainable
E-Commerce Logistics

Comparative analysis of ARIMA and Prophet using Amazon India sales data — 128,975 transactions, validated against the Global Superstore dataset, resolving exactly when each model wins.

0
Transactions Analyzed
0
Datasets Validated
0
Forecast Error Reduction
0
Estimated Annual Savings
LIVE MODEL COMPARISON Prophet Recommended
37.47
Prophet MAE (units/day)
54.10
ARIMA MAE (units/day)
30.74%
Accuracy Improvement
14
Day Forecast Horizon
Executive Summary

Three questions. One data-driven answer.

The research was designed around a single operating question for e-commerce leaders: which forecasting model should we trust, and when?

Research Question

Does Prophet's AI-driven seasonality decomposition outperform ARIMA's classical statistics for e-commerce demand — and does the answer change with portfolio size?

Key Findings

Prophet wins at 1–2 categories (30.74% lower MAE). ARIMA wins at 3+ categories. A precise "3-category tipping point" governs model choice.

Business Value

A tested decision framework that cuts forecast error by up to 32.76%, reduces inventory waste, and lowers the carbon footprint of logistics operations.

0
Accuracy Improvement (Category)
0
Estimated Annual Savings
0
Inventory Units Saved / Day
0
Peak Forecast Confidence
Research Problem

Two models. No shared guidance on when to use either.

ARIMA remains the industry default for stable, linear demand — but e-commerce is volatile, seasonal, and fast-moving. Prophet was purpose-built for exactly that pattern, yet practitioners have no data-driven rule for choosing between them.

Inaccurate model selection compounds into overstock, higher warehousing cost, product wastage, and a larger carbon footprint — the study set out to replace guesswork with a tested framework.

Primary Objective

A data-driven decision framework for model selection.

Implement ARIMA (1,1,1) and Prophet on 128,975 real Amazon India transactions using Python in Google Colab.

Evaluate both models with MAE and RMSE at category-level and total-demand level, then identify the category-count "tipping point."

Validate findings on the independent Global Superstore dataset and quantify the sustainability impact of optimal model choice.

Research Methodology

A reproducible pipeline, start to finish.

Quantitative, descriptive-comparative design. Secondary data, positivist paradigm, fully reproducible in Google Colab (Python 3.12).

STEP 1

Data Ingestion

128,975 raw transactions, 24 variables, Kaggle

STEP 2

Cleaning

Cancelled orders & nulls removed → 110,643 rows

STEP 3

Aggregation

Daily Qty by category, chronological split

STEP 4

Train / Test

77 days train / 14 days test

STEP 5

Model Fit

ARIMA(1,1,1) & Prophet (weekly seasonality)

STEP 6

Evaluation

MAE & RMSE via scikit-learn

Executive Dashboard Preview

The numbers leadership actually asks for.

Aggregate demand across the full nine-category Amazon India portfolio, March–June 2022.

91 days
110,643
Fulfilled Transactions
9 tracked
9
Product Categories
14-day
94%
Best-Case Forecast Accuracy
avg
395
Inventory Units Saved / Day
recommended
Category-Aware
Recommended Strategy
Forecast Comparison

Actual demand vs. ARIMA vs. Prophet.

14-day holdout test for the highest-volume "Set" category (42,947 records). Hover any point for the exact value.

P

Prophet — MAE 37.47 · RMSE 56.39

Weekly seasonality captures day-of-week shopping rhythm, keeping error 30.74% below ARIMA at single-category granularity.

A

ARIMA(1,1,1) — MAE 54.10 · RMSE 75.48

Predictions converge near the 77-day mean (~430 units/day), reflecting limited autocorrelation in volatile category-level data.

Business Insights

What the error curves actually mean for operations.

Four operational read-outs drawn directly from the category-level and aggregate-level results.

Prophet When Prophet Performs Better

At 1–2 product categories, demand carries strong weekly seasonality. Prophet's trend + seasonality decomposition captures this directly, cutting MAE by up to 30.74%.

ARIMA When ARIMA Performs Better

Once three or more categories are aggregated, individual volatilities cancel out. The smoother series favours ARIMA's linear trend modelling — a ~32% error reduction.

Operational Implications

A single "one model fits all" policy is measurably suboptimal. Portfolio size — not intuition — should decide which engine drives replenishment planning.

Inventory Recommendations

Run Prophet at SKU/category granularity for top sellers; run ARIMA at aggregate warehouse level. The hybrid approach captured the best of both in this study.

AI Recommendation Engine

What the model recommends, and why.

LIVE RECOMMENDATION · SET CATEGORY
Use Prophet
Lower MAE, superior weekly-seasonality capture, and a 96% forecast confidence score for single-category demand.
96%
Forecast Confidence
−30.74%
MAE vs ARIMA
≤2
Categories → Use Prophet
≥3
Categories → Use ARIMA
Business Impact

From forecast accuracy to bottom line.

Estimated impact of switching to the category-optimal model across the full portfolio.

0

Estimated annual inventory cost savings at 9-category total-demand optimisation.

%

0

Maximum forecast error reduction achieved at multi-category aggregate level.

U

0

Peak inventory units saved per day through optimal model selection.

Sustainability Analytics

Forecasting accuracy is an ESG lever.

Reduced forecast error translates directly into less overstock, fewer emergency shipments, and a smaller logistics carbon footprint.

0
Max Inventory Waste Reduction
0
Units Saved / Day (peak)
↓ CO₂e
Fewer Emergency Air Shipments
Warehouse Utilisation
Circular Economy Impact
ROI & Carbon Calculator

What would this be worth for your business?

Enter your own order volume and warehousing cost to estimate savings from switching to the category-optimal forecasting model — using the actual waste-reduction percentages measured in this study.

ARIMA
Recommended model at this category count
0
Estimated units saved per day
₹0
Daily cost saved
₹0
Annual cost saved
0 t
Annual CO₂e avoided (illustrative estimate)
Research Validation

Cross-dataset consistency, not a one-off result.

The identical pipeline was re-run on the independent Global Superstore dataset (2011–2014, 51,290 records, 3 categories).

Amazon India Dataset

128,975 records · Set category · Prophet MAE 37.47 vs ARIMA 54.10

Global Superstore Dataset

51,290 records · Office Supplies · Prophet MAE 30.56 vs ARIMA 51.37

Result Consistency ✓

Prophet wins at single-category level on both datasets — 40.51% advantage confirmed.

Decision Center

Boardroom-style executive recommendation.

A consulting-format summary of the full research, condensed for executive sign-off.

Business Problem
No shared, data-driven rule exists for choosing between ARIMA and Prophet across a multi-category e-commerce portfolio.
Research Findings
Prophet outperforms ARIMA by 30.74% at 1–2 categories; ARIMA outperforms Prophet by up to 32.76% at 3+ categories.
Recommended Model
Hybrid framework — Prophet for top 1–2 selling categories, ARIMA for aggregate/warehouse-level planning.
Operational Impact
Up to 395 units/day of inventory waste avoided; more reliable safety-stock planning via Prophet's uncertainty intervals.
Financial Impact
Estimated ₹71 lakhs in annual inventory cost savings at full 9-category optimisation.
Sustainability Impact
Lower overstock and fewer emergency shipments reduce warehousing energy use and transport-related carbon emissions.
Final Recommendation
Adopt the 3-category tipping point as standing forecasting policy.
Open Forecast Studio →
Model Recommender · 30-Second Quiz

Which model should you use?

Answer three questions about your own forecasting setup and get a data-backed recommendation, grounded in this study's actual results.

1. How many product categories do you need to forecast?

1–2 categories
3–5 categories
6+ categories

2. How volatile is your demand day-to-day?

Fairly stable
Moderately volatile
Highly volatile / spiky

3. What's your forecast horizon?

Short-term (< 14 days)
Medium-term (2–4 weeks)
Long-term (1+ months)
Recommended Forecasting Model
Explore Full Dashboard →
Interactive Dashboard · Forecast Studio

Explore the tipping point yourself.

Select a category count to see live MAE / RMSE comparisons and the recommended model update in real time.

What is a "category" here?

In e-commerce logistics analytics, a category is a product grouping used to aggregate SKU-level sales into a single demand time-series — e.g. "Fashion > Sets," "Fashion > Kurtas," or "Electronics." Warehouses and forecasting systems rarely predict demand item-by-item; instead they forecast at the category level, since that's the level at which inventory, shelf space, and replenishment decisions are actually made.

In this project, each category count on the slider below represents how many of these product categories were summed together into one daily demand series before forecasting — starting from a single category ("Set" alone, 42,947 records) and scaling up to all 9 categories in the Amazon Sale Report dataset combined. This is exactly what drives the result: forecasting one narrow, volatile category behaves very differently from forecasting nine categories pooled into one smooth aggregate signal — which is why Prophet wins at low category counts and ARIMA overtakes it at three or more.

Tipping Point Explorer
What-If Simulator
Complete Results Table
Model Configuration
ARIMA MAE
54.10
Prophet MAE
37.47

Hover over any bar to preview its numbers live; click a chip to lock that category count in.

Research Paper

The full academic report.

Seven chapters — literature review, methodology, full data analysis, conclusions, and recommendations — as submitted for the MBA (Artificial Intelligence & Data Science) capstone.

A Study on AI-Driven Demand Forecasting for Sustainable E-Commerce Logistics

Puvvada Rohan Sai Pavan · EA2452001011709 · SRM Institute of Science and Technology · May 2026

About the Researcher

Author

PR

Puvvada Rohan Sai Pavan

MBA — Artificial Intelligence & Data Science · SRM Institute of Science and Technology, Kattankulathur · Academic Year 2024–2026

Guided by Dr. Veena Christy, Assistant Professor. Independent research conducted end-to-end in Python (Google Colab) using publicly available Kaggle datasets.