This project aims to predict customer satisfaction in a delivery service based on survey responses. Using various machine learning techniques, the goal is to determine whether a customer is happy or not based on their feedback. This notebook showcases my ability to handle a real-world business problem, implement machine learning models, and evaluate their performance effectively.
Predict if a customer is happy or not based on their answers to specific survey questions.
The dataset contains the following attributes:
- Y: Target attribute indicating customer happiness (0 for unhappy, 1 for happy)
- X1: Order was delivered on time
- X2: Contents of the order met expectations
- X3: Ordered everything desired
- X4: Paid a good price for the order
- X5: Satisfaction with the courier
- X6: App ease of use
Attributes X1 to X6 represent responses ranging from 1 to 5, where smaller numbers indicate less satisfaction and higher numbers indicate more satisfaction.
The initial success metric is to achieve an accuracy score of 73% or higher. But after analyzing the business need, I concluded that predicting unhappy customers (Recall 0) will be better for the company since this client category needs to be forecasted first to take proactive actions to satisfy them.
- Data Preprocessing: Handling missing values, encoding categorical variables, and feature scaling.
- Model Training: Implementation and Comparison of multiple machine learning models to identify the best performer (Decision Tree, RandomForest, Logistic Regression, AdaBoost, XGBoost).
- Model Emprovement: : Using multiples techniaues for model improvement GridSearCV, RandomSearchCV, Ensemble learning, Voting techniques and Stacking.
- Evaluation Metrics: Calculation of accuracy, precision, recall (including specific labels), and F1-score.
- Cross-Validation: Utilization of cross-validation to ensure the robustness of the model.
- Visualization: Data visualization to understand the distribution and relationships within the data.
- Feature selection: Selecting the most important features (Only important questions for survey) using Recursive Feature Elimination (RFE) .
- Clone the repository:
git clone https://github.com/bellapp/dq5aU4G3aakwhO0a.git
- Navigate to the project directory:
cd dq5aU4G3aakwhO0a - Install the required packages:
pip install -r requirements.txt
- Run the Jupyter notebook:
jupyter notebook Happy_Customer_notebook.ipynb
The notebook demonstrates the step-by-step approach to solving the problem, from data preprocessing to model evaluation. The best model achieved an accuracy of 85% and a recall (0) of 92%. Detailed performance metrics and visualizations are provided to support the results.
The most important questions to put in the survey are :
- 'X1': my order was delivered on time
- 'X2': contents of my order was as I expected
- 'X5': I am satisfied with my courier