📊 Pharma Sales Forecasting: Time Series Analysis & Predictive Modeling

A comprehensive data science project for pharmaceutical sales forecasting using advanced machine learning techniques, featuring interactive dashboards and automated reporting.

---

📑 Table of Contents

• Project Overview
• Interactive Dashboard
• Business Problem
• Key Features
• Dataset Description
• Methodology
• Results & Insights
• Visualizations
• Project Structure
• Installation & Usage
• Technologies Used
• Future Improvements

---

🎯 Project Overview

This project addresses a critical challenge in the pharmaceutical industry: accurate sales forecasting across multiple drug categories. By leveraging time series analysis and machine learning, I developed predictive models to support New Product Development (NPD) and strategic decision-making.

Key Objectives:

✅ Forecast monthly pharmaceutical sales across 8 ATC drug categories
✅ Compare baseline (Prophet) vs. advanced (XGBoost) models
✅ Generate actionable insights for NPD strategy
✅ Create interactive dashboards for stakeholder presentations

---

🌐 Interactive Dashboard

📊 View Live Dashboard →

Click image to view interactive dashboard with all 8 drug categories

Dashboard Features:

• 🔄 Toggle between 8 drug categories (M01AB, M01AE, N02BA, N02BE, N05B, N05C, R03, R06)
• 📊 Compare Actual vs Prophet vs XGBoost predictions
• 🎯 Interactive hover tooltips showing exact values
• 🔍 Zoom and pan capabilities for detailed analysis
• 📱 Responsive design works on mobile and desktop

---

💼 Business Problem

Challenge:

Pharmaceutical companies need accurate sales forecasts to:

• Optimize inventory management
• Plan production schedules
• Identify high-growth product categories
• Support New Product Development decisions
• Allocate marketing resources effectively

Solution:

Developed a dual-model forecasting system combining:

1. Prophet (Facebook's time series tool) - Baseline model
2. XGBoost with feature engineering - Advanced model

---

✨ Key Features

🔍 Exploratory Data Analysis (EDA)

• Comprehensive statistical analysis of 2,106+ daily sales records
• Temporal pattern identification (trends, seasonality)
• Missing value analysis and data quality checks
• Interactive visualizations with Plotly

🛠️ Feature Engineering

• Lag Features: Historical sales values (1-month lag)
• Rolling Averages: 3-month moving averages
• Temporal Features: Year, Month extraction
• Category-Specific Features: Per-drug engineered variables

🤖 Advanced Modeling

• Baseline Model: Prophet with yearly seasonality
• Advanced Model: XGBoost with Optuna hyperparameter tuning
• Cross-Validation: Time-based train-test split (80/20)
• Evaluation Metrics: MAE, RMSE for model comparison

---

📦 Dataset Description

Source:

Pharmaceutical Sales Data (Kaggle)

Scope:

• Period: January 2014 - October 2019
• Granularity: Daily → Aggregated to Monthly
• Records: 2,106 daily entries → 70 monthly aggregates

ATC Drug Categories:

Category	Description	Avg Monthly Sales
M01AB	Anti-inflammatory (non-steroids)	5.03
M01AE	Propionic acid derivatives	3.90
N02BA	Salicylic acid derivatives	3.88
N02BE	Anilides (Pain relievers)	29.92 ⭐
N05B	Anxiolytics	8.85
N05C	Hypnotics and sedatives	0.59
R03	Anti-asthmatics	5.51
R06	Antihistamines	2.90

Note: N02BE (Pain relievers) shows highest sales volume - prime candidate for NPD focus.

---

🔬 Methodology

1. Data Preprocessing

# Date conversion & resampling to monthly frequency
df['datum'] = pd.to_datetime(df['datum'])
df_monthly = df.resample('ME').sum()

# Feature engineering
df['lag1'] = df['sales'].shift(1)
df['ma3'] = df['sales'].rolling(3).mean()

2. Model Development

Prophet Model (Baseline)

• Seasonality: Yearly patterns enabled
• Training: 54 months of historical data
• Forecasting: 14-month test period
• Best For: Capturing trend + seasonality

XGBoost Model (Advanced)

• Hyperparameter Tuning: Optuna (20 trials)
• Features: Lags + Moving Averages
• Optimization Target: Minimize MAE
• Best For: Capturing complex non-linear patterns

3. Evaluation Framework

MAE = mean_absolute_error(actual, predicted)
RMSE = sqrt(mean_squared_error(actual, predicted))

---

📈 Results & Insights

Model Performance Comparison

N02BE Category (Pain Relievers)

Model	MAE	RMSE	Performance
Prophet	313.74	434.05	Baseline
XGBoost	163.80 ⭐	208.06 ⭐	47.8% better

Key Finding: XGBoost significantly outperforms Prophet due to feature engineering capturing recent sales patterns.

Feature Importance Analysis

Top features driving N02BE sales predictions

Top Predictors (across all categories):

1. Lag Features (previous month sales) - 45% importance
2. 3-Month Moving Average - 30% importance
3. Seasonal Components - 25% importance

Business Recommendations

For New Product Development Team:

1. 📍 Priority Categories:

   • N02BE (Pain Relievers): Highest volume + stable demand
   • N05B (Anxiolytics): Growing trend detected
   • R03 (Anti-asthmatics): Seasonal peaks exploitable

2. 📊 Inventory Optimization:

   • Use XGBoost forecasts for procurement planning
   • Stock 10-15% buffer for high-variance categories (N05C)

3. 🎯 Marketing Strategy:

   • Align campaigns with predicted demand peaks
   • Focus resources on N02BE category expansion

4. 🔮 Future Model Enhancements:

   • Integrate external data (holidays, promotions)
   • Implement ensemble methods (Prophet + XGBoost)
   • Add competitor pricing features

---

📊 Visualizations

Sales Forecasting Results

Actual vs Predicted sales for N02BE category showing XGBoost superior performance

Key Insights:

• ✅ XGBoost captures short-term fluctuations better than Prophet
• ✅ Both models identify seasonal patterns successfully
• ✅ Prophet tends to over-predict during low-demand periods
• ✅ XGBoost maintains accuracy across volatile market conditions

---

📂 Project Structure

pharma-sales-forecasting/
│
├── pharmasales.ipynb             # Main analysis notebook
├── sales_dashboard.html          # Interactive dashboard
├── Sales_N02BE.png              # Results visualization
├── XGboot_feature_N02BE.PNG     # Feature importance chart
├── requirements.txt              # Python dependencies
├── README.md                     # This file
├── LICENSE                       # Project license
└── .gitignore                    # Git exclusions

---

🚀 Installation & Usage

Prerequisites

• Python 3.11+
• pip package manager
• Jupyter Notebook

Step 1: Clone Repository

git clone https://github.com/Omneya21/Pharma-Sales-Forecasting.git
cd Pharma-Sales-Forecasting

Step 2: Install Dependencies

pip install -r requirements.txt

Step 3: Run Analysis

# Open Jupyter Notebook
jupyter notebook pharmasales.ipynb

Step 4: View Interactive Dashboard

Option 1: View Online (Recommended)

https://htmlpreview.github.io/?https://github.com/Omneya21/Pharma-Sales-Forecasting/blob/main/sales_dashboard.html

Option 2: View Locally

# Simply double-click sales_dashboard.html
# Or open in browser:
firefox sales_dashboard.html  # Linux
open sales_dashboard.html     # Mac
start sales_dashboard.html    # Windows

---

🛠️ Technologies Used

Core Libraries:

• pandas (2.0+) - Data manipulation
• numpy (1.24+) - Numerical computing
• matplotlib / seaborn - Static visualizations
• plotly - Interactive dashboards

Machine Learning:

• Prophet (1.1+) - Time series baseline
• XGBoost (2.0+) - Gradient boosting
• scikit-learn (1.3+) - Model evaluation
• Optuna (3.0+) - Hyperparameter optimization

Development:

• Jupyter Notebook - Interactive analysis
• Git - Version control
• Python - Primary language

---

🔮 Future Improvements

Model Enhancements:

• Implement ensemble methods (stacking Prophet + XGBoost)
• Add LSTM/GRU for deep learning comparison
• Integrate external features (holidays, weather, economic indicators)
• Develop multi-step ahead forecasts (3-6 months)

Feature Engineering:

• Add competitor pricing data
• Include promotional campaign indicators
• Incorporate seasonality index per category
• Create product lifecycle stage features

Infrastructure:

• Deploy model as REST API (FastAPI)
• Automate retraining pipeline (MLflow)
• Build real-time dashboard (Streamlit)
• Implement CI/CD with GitHub Actions

Business Extensions:

• Add demand forecasting for new product launches
• Create what-if scenario analysis tool
• Develop ROI calculator for NPD decisions