Build a binary classification model to predict whether a customer will churn (i.e., discontinue their subscription) or not based on various features in the dataset. You are given a dataset containing customer information and whether they churned or not in the past. Your task is to use this dataset to train a machine learning model and evaluate its performance on a test set.
Clone the repository and install the required packages using pip:
git clone https://github.com/mohammadrezashariat/Churn-Prediction
cd Churn-Prediction
pip install -r requirements.txt
To run the prediction model, execute the following command in the project root directory:
python main.py
This will preprocess the data, train the model, and generate predictions for the test set. The model accuracy will be printed to the console.
You can use the Telco Customer Churn dataset, which is available on Kaggle at https://www.kaggle.com/blastchar/telco-customer-churn.
The prediction model is a logistic regression classifier, trained using scikit-learn.
The dataset was preprocessed using the following steps:
There were no duplicate rows in the dataset, so no rows were removed.
The "customerID" column was removed as it does not contribute to predicting customer churn.
There were missing values in 'TotalCharges' the dataset, so fill with mean of 'TotalCharges'.
Categorical variables such as :
['gender', 'InternetService', 'PaymentMethod', 'Partner', 'Dependents', 'PhoneService', 'PaperlessBilling', 'MultipleLines', 'OnlineSecurity', 'OnlineBackup', 'DeviceProtection', 'TechSupport', 'StreamingTV', 'StreamingMovies', 'Contract', 'PaperlessBilling']
were converted to numerical using one-hot encoding.
The numerical variables "MonthlyCharges", and "TotalCharges" were scaled using np.log1p.
The features used for training the model were selected based on their relevance in predicting customer churn. The selected features were
["gender", "SeniorCitizen", "Partner", "Dependents", "tenure", "PhoneService", "MultipleLines", "InternetService", "OnlineSecurity", "OnlineBackup", "DeviceProtection", "TechSupport", "StreamingTV", "StreamingMovies", "Contract", "MonthlyCharges",
"TotalCharges"]
No additional feature engineering was performed.
'Logistic regression' was chosen as the algorithm for this problem because it is a simple yet effective algorithm for binary classification problems such as predicting customer churn.
The model was evaluated using the following metrics:
Accuracy: The proportion of correct predictions made by the model.
ROC Curve: A plot of the true positive rate (TPR) against the false positive rate (FPR) for different classification thresholds.
Confusion Matrix: A matrix showing the number of true positives, true negatives, false positives, and false negatives predicted by the model.
