Webinar Engagement Lead Scoring Model

This project analyzes webinar engagement data for a fictional company, ProfDev, which runs weekly professional-development webinars. The goal is to identify high-potential leads who are most likely to convert into a business opportunity, enabling more efficient and targeted sales outreach.

The analysis focuses on behavioral engagement signals such as webinar attendance, duration, survey responses, and geographic attributes to predict whether a lead will convert.

---

Project Overview

As part of a marketing analytics and business operations workflow, this project builds a supervised classification model to predict lead conversion (i.e., whether a user books a demo).

The workflow includes:

• Data cleaning and preprocessing
• Exploratory data analysis (EDA)
• Feature engineering and encoding
• Handling class imbalance
• Model training (Logistic Regression)
• Model evaluation using multiple metrics beyond accuracy
• Feature interpretation and business insight generation

A key focus of this project is addressing class imbalance, which initially resulted in misleadingly high accuracy but poor predictive performance for the minority class.

---

Dataset Description

Each row represents a unique lead interacting with ProfDev’s webinar funnel.

Engagement Metrics

• Number of webinar registrations
• Number of webinars attended
• Average webinar attendance duration
• Number of sessions with >50% attendance duration

Survey Behavior

• Post-webinar survey response indicating interest in speaking with the team
  • 1 = expressed interest
  • 0 = no interest / no response

Target Variable

• Opportunity (binary classification target)
  • 1 = booked a demo / converted
  • 0 = did not convert

---

Modeling Approach

Baseline Model

A Logistic Regression model was trained to establish a baseline for conversion prediction.

Class Imbalance Handling

The dataset exhibited significant class imbalance, with far fewer positive conversions than non-conversions. To address this, oversampling (upsampling the minority class) was applied to improve model learning and recall for the positive class.

Evaluation Strategy

Model performance was evaluated using:

• Accuracy (baseline reference only)
• Confusion Matrix
• Precision, Recall, and F1-score
• ROC Curve and AUC
• Precision-Recall Curve (important for imbalanced classification)

---

Key Findings

• Survey engagement (expressed interest in speaking with ProfDev) is the strongest predictor of conversion.
• Webinar engagement behaviors (attendance frequency and duration) also contribute meaningfully to prediction performance.
• Geographic features were less predictive and risk introducing overfitting or reduced generalization.
• The initial high accuracy masked poor minority-class performance, highlighting the importance of using recall, ROC-AUC, and PR curves for evaluation.

---

Repository Structure

.
├── market_analysis.ipynb
├── data/
├── README.md