The Zomato Data Analysis Project is designed to extract actionable insights from Zomato's restaurant data using Azure cloud services. By implementing robust ETL (Extract, Transform, Load) pipelines, this project analyzes customer preferences, restaurant ratings, and ordering patterns to provide business intelligence for restaurant owners and stakeholders.
This project utilizes a modern data engineering and analytics stack, leveraging Azure services and Python libraries for efficient data processing and visualization:
Data is ingested from Azure Blob Storage, processed using PySpark in Databricks, analyzed with machine learning models, and visualized using Matplotlib to derive insights. The architecture ensures scalability and fault tolerance.
Data was extracted from CSV files stored in Azure Blob Storage. PySpark was used to clean the dataset, handling missing values, duplicates, and inconsistencies through imputation, deduplication, and outlier detection.
Built scalable ETL pipelines in Azure Databricks. The pipelines included:
Developed and evaluated multiple models to predict restaurant popularity and ratings:
Models were evaluated using RMSE, with Gradient Boosting achieving the lowest error rate.
Used Matplotlib and Pandas to create visualizations, including bar charts for cuisine preferences, scatter plots for rating distributions, and line graphs for spending trends.
North Indian and Chinese cuisines dominated, accounting for 65% of orders, with fast food and desserts also popular among younger demographics.
Restaurants with ratings above 4.2 stars, emphasizing the role of online reputation in customer loyalty.
Average spending per order was ₹800 for couples, with fine dining restaurants showing 25% higher repeat bookings.
Fast food and desserts had 70% of orders placed online, while fine dining saw stronger offline engagement, indicating distinct marketing needs.
The insights drove:
Challenge: Inconsistent data with missing values, duplicates, and outliers.
Solution: Applied PySpark-based cleaning, including median imputation for missing values, deduplication algorithms, and z-score-based outlier removal.
Challenge: Large datasets caused slow processing and high compute costs.
Solution: Optimized pipelines with data partitioning, caching, and selective column loading, reducing processing time by 45%.
Challenge: Initial models had high variance and low predictive accuracy.
Solution: Used feature engineering (e.g., cuisine type encoding) and ensemble models, improving RMSE by 30%.
This project delivered a robust data analysis solution for Zomato, leveraging Azure Databricks and machine learning to uncover insights into customer behavior and restaurant performance. The ETL pipelines and models are scalable and adaptable for other datasets.
The insights empowered restaurants to optimize menus, refine pricing, and enhance customer engagement, driving measurable business growth. The methodology serves as a blueprint for future data-driven projects in the food industry.