Churn Prediction for SaaS using Machine Learning

This project aims to build and compare several machine learning models (Softmax Regression, SVM, Random Forest) to predict a customer's churn risk score (1-5) and identify the key drivers of churn in a SaaS business context.

---

Project Objective

The goal is to develop a predictive model that can:

• Predict the churn risk score (1-5) for each customer.
• Identify the most important features influencing churn.
• Compare the performance of multiple machine learning algorithms.

---

Dataset Description

• The dataset contains customer information such as age, gender, region, membership category, joining date, referral status, transaction history, and feedback.
• The target variable is churnriskscore, which ranges from 1 to 5, representing the likelihood of churn.
• Key features include:
  • age, gender, regioncategory, membershipcategory
  • avgtransactionvalue, avgfrequencylogindays, pointsinwallet
  • usedspecialdiscount, offerapplicationpreference, pastcomplaint
  • feedback (categorical, processed using NLP techniques)

Dataset Used: Churn Risk Dataset on Kaggle

---

Methodology

1. Data Preprocessing

   • Handle missing values and drop irrelevant columns (e.g., lastvisittime).
   • Convert categorical features to numerical using encoding.
   • Use NLP to process textual feedback into numerical features.

2. Exploratory Data Analysis (EDA)

   • Visualize the distribution of churn risk scores.
   • Analyze correlations between features and churn.

3. Feature Engineering

   • Extract meaningful features from raw data.
   • Use SMOTE to balance the training dataset.

4. Model Selection & Training

   • Train and compare Softmax Regression, SVM, and Random Forest models.
   • Perform hyperparameter tuning using Optuna for Random Forest.

5. Model Evaluation

   • Evaluate models using accuracy, precision, recall, and F1-score.
   • Use cross-validation for robustness.

---

Model Comparison

Model	Accuracy	Precision	Recall	F1-Score
Softmax Regression	0.58	0.56	0.58	0.55
SVM	0.73	0.73	0.73	0.72
Random Forest	0.74	0.74	0.74	0.74

---

Key Features

• NLP Integration: Feedback text is processed using lemmatization and stopword removal.
• Hyperparameter Tuning: Optuna is used for optimizing Random Forest hyperparameters.
• Imbalanced Data Handling: SMOTE is applied to balance the training dataset.
• Comprehensive Evaluation: Multiple metrics are used to assess model performance.

---

Results

• The Random Forest model achieved the highest accuracy (74%) and F1-score.
• Key drivers of churn include avgtransactionvalue, avgfrequencylogindays, and feedback.
• The model can be used to identify at-risk customers and implement targeted retention strategies.