Churn Prediction for Homelessness Prevention

Project Overview: A Canadian NGO is dedicated to providing housing first programs to homeless individuals and offering essential support to help them reintegrate into stable living conditions. However, one of their major challenges was the unexpected dropout of participants from the program. To address this issue, ByteWise partnered with the NGO to develop a predictive solution using machine learning techniques. The goal was to accurately identify individuals at risk of leaving the program, allowing the NGO to proactively intervene and provide the necessary support.

Challenges:

1. High Dropout Rates: The NGO faced a significant issue with program participants unexpectedly dropping out, hindering the effectiveness of their homelessness prevention initiatives.

2. Data Complexity: The available data included case worker notes, which varied in format and content, making it challenging to extract meaningful insights.

3. Data Cleaning and Imputation: The data required extensive cleaning and imputation, as missing values and inconsistencies were common. This added complexity to the model development process.

4. Class Imbalance: The dataset was imbalanced, with a limited number of dropout cases compared to those who remained in the program. Advanced techniques were necessary to address this imbalance.

5. Proactive Intervention: The NGO needed a tool to predict potential dropouts with enough lead time to intervene effectively, preventing participants from leaving the program prematurely.

Solution:

1. Data Collection and Integration: ByteWise collaborated with the NGO to collect and integrate historical data, including case worker notes, participant demographics, and program engagement information.

2. Data Preprocessing: Our data scientists performed extensive data cleaning and imputation to address missing values and inconsistencies, ensuring the dataset's quality.

3. Balancing the Data: Advanced techniques, such as synthetic data generation, were applied to address the class imbalance within the dataset.

4. Feature Engineering: Our team utilized advanced feature engineering techniques to process the unstructured case worker notes, extracting valuable insights and sentiment analysis for predictive features.

5. Machine Learning Model Development: ByteWise developed and tested several machine learning models, including Support Vector Classifiers, K-Nearest Neighbors, and Logistic Regression to accurately predict the likelihood of program dropout.

6. Model Evaluation: The models were rigorously tested using historical data, and the one with the highest accuracy was selected for deployment.

7. Deployment and Integration: The chosen model was seamlessly integrated into the NGOs existing data infrastructure, enabling real-time prediction and intervention.

Results:

• 81% Accuracy: The machine learning model achieved an accuracy rate of 81% in predicting program dropout, providing the NGO with a powerful tool for proactive intervention.

• Early Intervention: With the model in place, the NGO was able to identify at-risk individuals and allocate resources to offer timely support, significantly reducing unexpected dropouts.

• Cost Savings: By preventing participants from prematurely leaving the program, the NGO saw reduced operational costs and more efficient resource allocation.

Conclusion:

ByteWise's data-driven approach empowered the NGO to address the challenge of program dropout effectively. By leveraging machine learning, advanced data cleaning and imputation, as well as addressing class imbalance, the organization now has the tools to proactively support program participants, resulting in improved outcomes and reduced homelessness rates. This project stands as a testament to the power of data analytics in driving positive social impact and transforming the lives of vulnerable individuals. If you are looking for ways to leverage your data, contact ByteWise today and let us help you make a difference.