Startup Fundraising Analysis (Crunchbase Dataset)

Project: 2
Difficulty: Beginner
Tools: SQLite, Juypter Notebook, Python, Focus: Data Cleaning, Analysis, Using SQLite through a Python import, using pandas import.

Project Overview

This project explores startup investment data from the Crunchbase dataset using a combined pandas + SQLite workflow. The objective was to transform raw, unstructured data into a clean, queryable format and derive insights into fundraising trends, sector performance, and funding patterns.

The project follows a full data pipeline:

• Data ingestion from CSV
• Data cleaning and transformation
• Storage in a relational database (SQLite)
• Analytical querying using SQL
• Interpretation of business insights

---

Tools & Technologies

• Python
• Pandas
• SQLite
• Jupyter Notebook

---

Dataset

• Source: Crunchbase Startup Investments dataset

• Records:

  • Raw dataset: ~54,294 rows
  • Cleaned dataset: 40,907 rows

---

Data Processing Workflow

1. Data Exploration

• Inspected dataset structure, data types, and missing values

• Identified key issues:

  • Funding amounts stored as text
  • Date fields in string format
  • Inconsistent column naming
  • Missing and placeholder values

---

2. Data Loading (SQLite Integration)

• Loaded CSV into SQLite using chunk-based processing

• Prevented duplication by using:

  • replace for the first chunk
  • append for subsequent chunks

• Created a structured relational table for querying

---

3. Data Cleaning & Transformation

Key transformations included:

• Renaming inconsistent columns (e.g. _funding_total_usd_ → funding_total_usd)
• Converting funding values to numeric format
• Handling invalid placeholders (e.g. '-', empty values)
• Converting date columns to datetime
• Standardising founded_year to integer
• Removing records with missing or invalid funding values

Result:

• ~25% of records removed due to poor data quality

---

Key Analysis & Insights

1. Funding Trends Over Time

• Startup activity increased significantly from the early 2000s
• Peak funding observed around 2010–2012
• Average funding per company decreased over time, indicating broader distribution of capital

---

2. Most Successful Startup Verticals

• Biotechnology received the highest total funding
• Mobile had the highest average funding per company
• Software had many companies but lower average funding, indicating a competitive market

---

3. Funding Round Distribution

• Majority of startups received only 1–2 funding rounds
• Sharp drop-off in later rounds
• Indicates high attrition beyond early-stage funding

---

4. Funding Type Analysis

• Venture capital dominated total funding
• Private equity and debt financing contributed significantly at later stages
• Early-stage funding (seed, angel) occurred frequently but with lower capital amounts

---

5. Investment Concentration

• Funding is highly concentrated among a small number of companies
• Top firms received disproportionately large amounts of capital
• Indicates unequal distribution of investment across the startup ecosystem

---

Challenges & Limitations

1. Data Quality Issues

• Missing and inconsistent values required significant cleaning
• Funding column stored as text with symbols and placeholders
• Category fields were not fully normalised

2. Database Duplication Issue

• Initial data loading caused duplicate records due to repeated append operations

• Resolved by:

  • Rebuilding the database from scratch
  • Implementing controlled chunk loading (replace + append logic)

3. Lack of Investor-Level Data

• Dataset does not include investor names

• Prevented direct analysis of investor activity

• Adapted approach by analysing:

  • Funding types
  • Company-level investment patterns

4. Outliers in Funding Data

• Extremely large funding values skewed averages
• Required careful interpretation of results

Key Skills Demonstrated

• Data cleaning and preprocessing
• SQL querying and aggregation
• Database management (SQLite)
• Handling large datasets with chunking
• Data validation and debugging
• Translating data into business insights

Conclusion

This project demonstrates an end-to-end data analysis workflow, from raw data ingestion to actionable insights. It highlights real-world challenges such as data quality issues, duplication errors, and dataset limitations, while showcasing the ability to adapt analysis techniques accordingly.

Future Improvements

• Normalise category fields for more accurate sector analysis
• Incorporate additional datasets with investor-level detail
• Apply visualisation tools (e.g. Power BI, Tableau) for dashboard reporting