Sales Forecast AI

A web-based application for sales forecasting and analysis using machine learning. Built with Streamlit and Facebook's Prophet forecasting model, this tool helps businesses predict future sales trends based on historical data.

Features

Core Functionality

• Data Upload: Upload CSV files with historical sales data
• Sales Forecasting: Generate predictions using Facebook's Prophet model
• Advanced Analysis: Deep dive into sales patterns and seasonality
• Interactive Visualizations: Charts and graphs using Plotly
• Export Results: Download forecasts and analysis as CSV or PDF

Analysis Capabilities

• Trend Analysis: Identify long-term growth patterns
• Seasonality Detection: Uncover weekly, monthly, and yearly patterns
• Anomaly Detection: Find unusual sales patterns
• What-if Scenarios: Test different forecasting parameters
• Performance Metrics: MAE, RMSE, MAPE, and MSE calculations

Visualization Features

• Forecast Plots: Interactive charts with confidence intervals
• Component Analysis: Break down trend, seasonality, and holiday effects
• Residual Analysis: Model performance visualization
• Monthly Breakdowns: Detailed period analysis
• Seasonal Decomposition: Advanced time series analysis

Installation

Prerequisites

• Python 3.7 or higher
• pip (Python package installer)

Setup Instructions

1. Clone or download the project

   cd SalesForecastAI

2. Install required dependencies

   pip install -r requirements.txt

3. Run the application

   streamlit run app.py

4. Access the app

   • The application will automatically open in your default web browser
   • Default URL: http://localhost:8501

Data Requirements

Required Format

Your CSV file must contain at least two columns:

Column	Format	Example
Date	YYYY-MM-DD	2023-01-01
Sales	Numeric	1000.50

Optional Columns

You can include additional numeric columns that might influence sales:

• Marketing spend
• Price changes
• Promotional events
• External factors

Data Quality Tips

• Minimum Data: At least 1 year of historical data for better seasonality detection
• Consistent Format: Ensure dates are in YYYY-MM-DD format
• Clean Data: Remove or explain outliers
• Regular Intervals: Daily data works best, but weekly/monthly is also supported

Configuration

Model Parameters

The application uses Facebook's Prophet model with configurable parameters:

Parameter	Description	Default	Impact
changepoint_prior_scale	Trend flexibility	0.05	Higher = more flexible trend
seasonality_prior_scale	Seasonality strength	10.0	Higher = stronger seasonality
confidence_interval	Prediction uncertainty	0.95	95% confidence interval
forecast_period	Days to predict	365	One year forecast

Seasonality Settings

• Yearly Seasonality: Enable for annual patterns
• Weekly Seasonality: Enable for weekly patterns
• Daily Seasonality: Enable for daily patterns

Understanding Results

Forecast Components

• Trend: Long-term growth or decline patterns
• Seasonality: Recurring patterns (weekly, monthly, yearly)
• Holidays: Special event effects
• Confidence Intervals: Uncertainty ranges around predictions

Performance Metrics

• MAE (Mean Absolute Error): Average absolute prediction error
• RMSE (Root Mean Squared Error): Square root of average squared errors
• MAPE (Mean Absolute Percentage Error): Percentage error representation
• MSE (Mean Squared Error): Average squared prediction errors

Interpretation Tips

• Lower metric values indicate better model performance
• Wider confidence intervals suggest higher uncertainty
• Strong seasonality shows clear recurring patterns
• Anomalies may indicate special events or data issues

Technical Architecture

Frontend

• Streamlit: Web application framework
• Plotly: Interactive visualizations
• Custom CSS: Styled user interface

Backend

• Prophet: Facebook's forecasting library
• Pandas: Data manipulation and analysis
• NumPy: Numerical computations
• Scikit-learn: Additional machine learning utilities

Project Structure

SalesForecastAI/
├── app.py                 # Main application file
├── requirements.txt       # Python dependencies
├── components/           # UI components
│   ├── data_upload.py   # Data upload functionality
│   ├── forecast.py      # Forecasting interface
│   ├── analysis.py      # Analysis features
│   ├── help.py          # Help documentation
│   └── sidebar.py       # Settings sidebar
├── utils/               # Utility functions
│   ├── data_processing.py  # Data validation and processing
│   ├── forecasting.py      # Prophet model wrapper
│   ├── metrics.py          # Performance calculations
│   └── visualization.py    # Chart creation
└── styles/              # Custom styling
    └── custom.py        # CSS styles

Troubleshooting

Common Issues

Installation Problems

# If you encounter dependency issues:
pip install --upgrade pip
pip install -r requirements.txt --force-reinstall

Data Upload Errors

• Ensure CSV format is correct
• Check date format (YYYY-MM-DD)
• Verify sales column contains numeric values
• Remove any empty rows or invalid data

Forecast Generation Issues

• Ensure sufficient historical data (at least 1 year recommended)
• Try adjusting model parameters in sidebar
• Check for data quality issues or outliers

Performance Issues

• Reduce forecast period for faster processing
• Simplify model parameters
• Use smaller datasets for testing

Best Practices

For Better Forecasts

1. Provide More Data: Longer history improves accuracy
2. Clean Your Data: Remove outliers and errors
3. Include Relevant Variables: Add factors that influence sales
4. Regular Updates: Refresh forecasts with new data
5. Validate Results: Compare predictions with actual outcomes

For Data Preparation

1. Consistent Formatting: Use standardized date formats
2. Handle Missing Data: Fill gaps or remove incomplete records
3. Document Changes: Note any business events or anomalies
4. Quality Checks: Verify data accuracy before uploading

Pictures