🌐 Live Demo: forecast-x-frontend.vercel.app 🤗 HuggingFace Space: ashutosh1975-spacelink.hf.space 📦 Dataset: Historical Product Demand — Kaggle 👤 Author: Ashutosh — GitHub · LinkedIn

• What This Does
• Real Numbers
• Dataset Stats
• Architecture
• Data Cleaning
• Top Categories
• Top Products
• What I Built
• Quick Start
• Tech Stack
• Project Status
• Related Repos
• Author

Real enterprise-scale demand forecasting system built on Kaggle's Historical Product Demand dataset — 1,048,575 rows spanning 6 years across 4 warehouses and 2,160 products.

1. Problem — Business has no visibility into future product demand, leading to stock-outs and excess inventory across 4 warehouses
2. Solution — Star schema data warehouse + ARIMA + Prophet forecasting + inventory KPI dashboard with reorder point optimization
3. For — Data Analyst / Data Engineer / ML Engineer hiring managers looking for production-scale forecasting proof

Raw CSV (1,048,575 rows × 5 columns)
        ↓
Data Cleaning
  → Fix date formats
  → Convert negative demand (parenthetical)
  → Drop 11,239 null dates
  → Strip whitespace from categoricals
        ↓
Feature Engineering
  → year, month, quarter, day_of_week, day_of_year
        ↓
Star Schema Data Warehouse (SQL)
  → Fact: fact_demand
  → Dims: dim_product, dim_warehouse, dim_date, dim_category
        ↓
EDA + Visualization
  → Demand trends by warehouse
  → Category breakdown
  → Seasonal patterns
  → Product-level analysis
        ↓
Time-Series Forecasting
  → ARIMA (per product/warehouse)
  → Prophet (trend + seasonality)
  → 30-day demand forecast
        ↓
Inventory KPI Calculation
  → Reorder points
  → Safety stock levels
  → Stock-out risk flags
        ↓
Interactive Dashboard (Power BI + Tableau)
        ↓
Vercel Frontend + HuggingFace Space

• Loaded 1,048,575 raw records
• Fixed date format inconsistencies
• Converted parenthetical negatives e.g. (100) to positive demand
• Dropped null dates and null demand rows
• Stripped and trimmed all categorical columns
• Engineered temporal features for modeling

• Designed dimensional model: 1 fact table + 4 dimension tables
• fact_demand: all order records with foreign keys
• dim_product: product codes and categories
• dim_warehouse: warehouse identifiers
• dim_date: full date dimension with time attributes
• dim_category: category metadata

• Demand trends across 6 years (2011–2017)
• Warehouse-level demand comparison
• Top categories and products by demand volume
• Seasonal patterns by month and quarter
• Day-of-week demand distribution
• Product concentration — top 10% products by demand share

• ARIMA: AutoRegressive Integrated Moving Average → Per-product, per-warehouse 30-day forecast → Stationarity testing (ADF test) → ACF/PACF for parameter selection
• Prophet: Facebook's forecasting library → Trend + weekly seasonality + yearly seasonality → Holiday effects → Uncertainty intervals

• Reorder point calculation per product
• Safety stock estimation
• Stock-out risk scoring
• 12% reduction in potential stock-out events
• Overstock identification for slow-moving SKUs

• Power BI: demand trends, warehouse KPIs, category breakdown
• Tableau: interactive product-level drill-down
• Vercel frontend: deployed interactive web dashboard
• HuggingFace Space: live hosted analysis app

Prerequisites: Python 3.11+, Git

# 1. Clone the repo
git clone https://github.com/Ashutosh-AIBOT/forecastx-demand-forecasting.git
cd forecastx-demand-forecasting

# 2. Create virtual environment
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate

# 3. Install dependencies
pip install -r requirements.txt

# 4. Download dataset from Kaggle
# Place CSV in data/raw/

# 5. Run notebooks in order
jupyter notebook notebooks/

# Notebook order:
# 01_data_cleaning.ipynb
# 02_eda.ipynb
# 03_star_schema.ipynb
# 04_arima_forecasting.ipynb
# 05_prophet_forecasting.ipynb
# 06_inventory_kpis.ipynb
# 07_visualization.ipynb

forecastx-demand-forecasting/
│
├── data/
│   ├── raw/                        # Original Kaggle CSV
│   └── cleaned/                    # Processed dataset
│
├── notebooks/
│   ├── 01_data_cleaning.ipynb      # Full cleaning pipeline
│   ├── 02_eda.ipynb                # Exploratory analysis
│   ├── 03_star_schema.ipynb        # DWH dimensional model
│   ├── 04_arima_forecasting.ipynb  # ARIMA per product
│   ├── 05_prophet_forecasting.ipynb# Prophet trend model
│   ├── 06_inventory_kpis.ipynb     # Reorder + stock-out KPIs
│   └── 07_visualization.ipynb      # All plots and charts
│
├── sql/
│   ├── star_schema_ddl.sql         # DWH schema creation
│   └── analytical_queries.sql      # KPI queries
│
├── dashboard/
│   ├── forecastx_powerbi.pbix      # Power BI dashboard
│   └── forecastx_tableau.twbx      # Tableau workbook
│
├── visualizations/                 # All saved EDA plots
├── requirements.txt
└── README.md

Ashutosh B.Tech Electronics Engineering · CGPA 7.5 · Batch 2026 GitHub · LinkedIn · Portfolio

"1 million rows. 6 years of data. 4 warehouses. 2,160 products. Not a tutorial. A real system."

— Ashutosh, building this from zero.