Forecasting-ML-App

R machine learning application that performs forecasting on pharmaceutical medicine sales data using information obtained form NHS (UK) General Practitioner (GP) datasets.

NHS Prescription Forecasting Dashboard

Overview

This Shiny dashboard application provides time series forecasting capabilities for NHS prescription data. It allows users to analyze and forecast prescription quantities and sales revenue using various forecasting models.

Features

• Interactive product selection
• Multiple forecasting models:
  • Auto (ETS)
  • Holt-Winters
  • TBATS
  • Auto ARIMA
  • Markov Chain Monte-Carlo (Prophet)
• Metrics analysis:
  • Sales Revenue
  • Quantity
• Visualization components:
  • Top 5 related products comparison
  • Time series decomposition
  • Forecast plots with confidence intervals
• Detailed data tables for both historical and forecast data

Prerequisites

The following R packages are required:

required_packages <- c(
  "devtools",
  "shinyWidgets",
  "data.table",
  "DT",
  "forecast",
  "ggplot2",
  "shinyjs",
  "plotly",
  "xts",
  "prophet",
  "tidyverse",
  "shinydashboard"
)

Installation

1. Clone the repository:

git clone https://github.com/Sabareh/Forecasting-ML-App
cd Forecasting-ML-App

2. Install required packages:

for(pkg in required_packages){
  if(!require(pkg, character.only = TRUE)){
    install.packages(pkg)
  }
}

3. Ensure you have the data file:

• productdb.rds should be in the root directory

Usage

1. Start the application:

shiny::runApp()

2. Using the dashboard:
   • Select a product from the dropdown menu
   • Choose a forecasting model
   • Select the metric (Sales Revenue or Quantity)
   • Toggle the ETS decomposition if needed
   • View the results in the various plots and tables

Data Structure

The application expects a productdb.rds file with the following columns:

• BNFNAME: Product name
• CHEMSUB: Chemical substance
• QUANTITY: Prescription quantity
• ACTCOST: Actual cost
• month: Date in YYYY-MM format

Features Breakdown

Product Selection

• Dropdown menu with autocomplete functionality
• Shows all available products in the NHS database

Forecasting Models

1. Auto (ETS)

   • Automatic exponential smoothing
   • Best for general forecasting

2. Holt-Winters

   • Triple exponential smoothing
   • Handles trend and seasonality

3. TBATS

   • Handles complex seasonality
   • Good for multiple seasonal patterns

4. Auto ARIMA

   • Automatic ARIMA model selection
   • Best for stationary time series

5. MCMC (Prophet)

   • Facebook's Prophet algorithm
   • Handles missing data and outliers

Visualizations

1. Top 5 Plot

   • Shows related products
   • Interactive plotly visualization

2. Forecast Plot

   • Shows predictions with confidence intervals
   • Highlights trends and patterns

3. Time Series Plot

   • Historical data visualization
   • Optional ETS decomposition

Data Tables

1. Forecast Data

   • Shows predicted values
   • Includes confidence intervals
   • Currency formatting for revenue

2. Historical Data

   • Shows raw data
   • Sortable and searchable

Styling

• Custom CSS in www/packt.css
• Uses Google Fonts:
  • Fanwood Text
  • Varela

Error Handling

• Validates input data
• Handles missing data gracefully
• Provides user feedback for errors

Performance Considerations

• Uses reactive programming
• Efficient data.table operations
• Optimized plotting with plotly

Contributing

Please read CONTRIBUTING.md for details on our code of conduct and the process for submitting pull requests.

License

This project is licensed under the MIT License - see the LICENSE.md file for details.

Acknowledgments

• NHS Digital for the prescription data
• R Shiny community
• Contributors to the various R packages used

Support

For support, please open an issue in the repository or contact [maintainer email].

Version History

• 1.0.0: Initial release
  • Basic forecasting functionality
  • Five forecasting models
  • Interactive visualizations