🛒 Retail Data Analytics & Prediction System

An end-to-end Data Analytics and Machine Learning project built using Python.

This project simulates a real-world retail sales workflow where raw sales data is:

• cleaned and transformed,
• analyzed for business insights,
• visualized using charts,
• and used for machine learning predictions.

The project demonstrates practical skills across:

• Data Engineering
• Data Analysis
• Data Visualization
• Machine Learning

---

📌 Project Workflow

Raw Sales Data
      ↓
Data Cleaning & Transformation
      ↓
Exploratory Data Analysis (EDA)
      ↓
Business Insights & Visualization
      ↓
Machine Learning Experiments
      ↓
Prediction System

---

📌 About the Dataset

The dataset used in this project is intentionally small but raw and unclean.
This project was built primarily for learning purposes to understand real-world data workflows including:

• data cleaning,
• preprocessing,
• exploratory analysis,
• visualization,
• and machine learning pipelines.

The focus was on handling messy real-world issues such as:

• missing values,
• duplicates,
• inconsistent formatting,
• and invalid data types,

rather than building a production-scale model with a large dataset.

📂 Project Files

File	Purpose
Cleaning.py	Cleans and transforms raw sales data
Analysis.py	Performs exploratory analysis and visualization
FailML.py	Initial ML attempt using static Revenue formula
Prediction.py	Final ML model for quantity prediction
sales_raw.csv	Raw messy dataset
sales_cleaned.csv	Cleaned dataset generated after preprocessing

---

🏗️ Phase 1 — Data Engineering (Cleaning.py)

The raw dataset intentionally contained multiple real-world data issues to simulate industry-style preprocessing challenges.

✅ Data Cleaning Performed

1. Duplicate Removal

Removed duplicate records based on Order_ID.

df.drop_duplicates(subset='Order_ID')

---

2. Missing Value Handling

Handled missing values in:

• City
• Price
• Quantity
• Date

Approaches used:

• Median imputation
• Group-based median transformation
• Row removal for invalid dates
• Default values where appropriate

---

3. Data Type Conversion

Converted:

• Date → datetime
• Price → numeric
• Quantity → numeric

Used:

errors='coerce'

to safely handle invalid values.

---

4. Text Standardization

Standardized categorical columns using:

str.lower()

This helped maintain consistency in:

• Product names
• Categories
• Cities

---

5. Handling Invalid Data

Removed:

• Negative prices
• Zero or invalid quantities

Applied business logic to ensure realistic data.

---

6. Special Handling of Mixed Quantity Values

The dataset contained text values like:

"three"

Instead of blindly coercing them into NaN, they were manually mapped:

'three' → 3

This preserved valid business information instead of losing data.

---

7. Feature Engineering

Created additional columns:

• Revenue = Price × Quantity
• Year
• Month
• Month_Name

These features were later used for analysis and visualization.

---

📊 Phase 2 — Data Analysis (Analysis.py)

Performed exploratory data analysis (EDA) to extract business insights from cleaned sales data.

---

📈 Questions Answered

✅ Top-Selling Products

Identified products generating the highest revenue.

Visualization:

• Bar Chart

---

✅ Revenue by City

Analyzed city-wise revenue contribution.

Visualization:

• Pie Chart

---

✅ Monthly Sales Trend

Studied monthly revenue patterns and seasonality.

Visualization:

• Line Chart

---

✅ Category Revenue Share

Compared contribution of each product category.

Visualization:

• Pie Chart

---

🔍 Additional Business Questions Explored

Beyond the initial requirements, additional business-oriented questions were analyzed:

✅ Which city buys the most electronics?

Used filtered category-level aggregation to identify strongest electronics market.

---

✅ Which category sells the highest quantity?

Analyzed quantity-based demand rather than only revenue.

This helped distinguish:

• high-value products vs
• high-demand products

---

✅ Which month generated the highest revenue?

Used grouped monthly aggregation to identify peak business periods.

---

💡 Business Insights Generated

Examples of insights extracted:

• Highest revenue-driving product
• Strongest performing city
• Peak sales month
• Highest contributing category

These insights simulate real business reporting workflows.

---

🤖 Phase 3 — Initial ML Attempt (FailML.py)

An initial machine learning attempt was made to predict Revenue.

However, during experimentation it was realized that:

Revenue = Price × Quantity

This is a direct mathematical formula, not a hidden pattern.

Using Linear Regression on a deterministic formula does not represent meaningful machine learning.

❌ Why This Was Not Ideal

• Revenue was already mathematically defined.
• The model was effectively learning multiplication.
• Predictions were not valuable from a business intelligence perspective.

This experiment was intentionally kept in the project to demonstrate:

• critical thinking,
• model evaluation,
• and understanding of when ML is appropriate.

---

🤖 Phase 4 — Final Machine Learning Model (Prediction.py)

A more meaningful ML problem was designed:

🎯 Goal

Predict Quantity using:

• Price
• Product
• Category
• City

---

🧠 ML Workflow

✅ Train-Test Split

Dataset split:

• 80% training
• 20% testing

Used:

train_test_split()

---

✅ Handling Categorical Data

Since ML models require numerical input:

Used:

• OneHotEncoder
• ColumnTransformer

to convert text categories into machine-readable features.

---

✅ Pipeline-Based Modeling

Implemented:

Pipeline()

to automate:

1. Encoding
2. Preprocessing
3. Model Training

This created a cleaner and production-style ML workflow.

---

📉 Model Evaluation

Evaluated model performance using:

RMSE (Root Mean Squared Error)

Measures average prediction error magnitude.

R² Score

Measures how well the model explains data variance.

---

📊 Visualization

Generated:

• Actual vs Predicted Quantity plot

This helped visually inspect model performance.

---

👤 User Input Prediction System

Added interactive prediction functionality where users can enter:

• Price
• Product
• Category
• City

and receive predicted quantity output.

---

🛠️ Technologies Used

• Python
• Pandas
• NumPy
• Matplotlib
• Scikit-learn

---

📚 Key Concepts Demonstrated

Data Engineering

• Data Cleaning
• Missing Value Handling
• Feature Engineering
• Data Transformation

Data Analysis

• EDA
• Aggregation
• Visualization
• Business Insights

Machine Learning

• Regression
• Encoding
• Pipelines
• Model Evaluation

---

🚀 Future Improvements

Possible future enhancements:

• Larger datasets
• Power BI dashboard
• Time-series forecasting
• Advanced ML models
• Deployment using Flask/Streamlit

---

👩‍💻 Author

Simran

This project was built as a hands-on learning project to understand real-world data workflows from raw data to prediction systems.