How it works
This site answers one simple question: which customers may leave soon, and what clues make the model think that?
The dashboard shows which customers may stop using a service. In data projects, this is often called churn. Here, the UI mostly says leaving because that is easier to understand.
It highlights customers most likely to leave, the monthly money connected to those customers, and the biggest reasons the model sees.
A high chance of leaving means the model found a customer who looks similar to customers who left in the past. It is a warning sign, not a promise.
The reasons chart shows which customer details mattered most to the model. It helps explain the prediction, but it does not prove one detail caused the customer to leave.
Python cleans each customer row. It turns numbers like monthly bill and months as customer into model-ready values. It also turns words like contract type and payment type into simple 0/1 columns.
The model gives each customer a chance of leaving from 0% to 100%. The app then labels that score as Low, Medium, or High risk.
The project uses logistic regression from scikit-learn. That means the model predicts one of two answers: the customer leaves or the customer stays.
It is used because it is quick, easier to explain than many models, and gives a percentage chance of leaving. That makes it a good first model for a clear business dashboard.
Airbyte Cloud moves the public customer file into the Neon database.
Airbyte does not train the model. Its job is to load the data. After that, Python cleans the data, trains the model, and saves dashboard-ready tables.
Airbyte reads the public customer file and loads it into Neon PostgreSQL.
The database keeps the original customer rows so Python has a clean starting point.
Python cleans column names, fixes missing values, and prepares useful customer details.
Python trains a model that estimates each customer's chance of leaving.
The pipeline saves tables for main numbers, customer predictions, and leaving reasons.
The dashboard shows the results, and the chatbot answers questions using those tables.
raw.customer_churnOriginal customer rows loaded by Airbyte.
analytics.customer_healthClean customer rows used by the dashboard.
analytics.churn_predictionsOne row per customer with chance of leaving and risk level.
analytics.kpi_summaryMain dashboard numbers.
analytics.model_evaluationCounts that explain the model score cards.
analytics.churn_driversCustomer details that most affected the model.