Week 5 Assignment – Supervised Learning (Regression)

Overview Of Week 5 Assignment

Goal:

Build a baseline regression model to predict a numeric target variable from your dataset. In your case:

• Target variable: CSAT Score (Customer Satisfaction Score)
• Features: Other numeric or encoded columns like Item_price, connected_handling_time, etc.

The assignment focuses on supervised learning using regression, splitting the data into training and testing sets, building a Linear Regression model, evaluating it with MAE, MSE, RMSE, Explained Variance, and visualizing predictions.

Step 1: Load Dataset

• Load your dataset (Customer_support_data.csv) using Pandas.
• Check the first 10 rows to understand the structure and columns.
• Identify the target column (CSAT Score) and feature columns (numeric predictors).

🔎 First 5 Rows (Preview)

Step 2: Preprocessing

• Drop high-cardinality ID columns: Customer_ID, Order_ID, Product_ID → These don’t help the model and may cause memory issues.
• Handle missing values: Fill missing numeric values with median.
• Encode categorical variables:
  • Only encode low-cardinality categorical columns (less than 20 unique values).
  • Use one-hot encoding (pd.get_dummies) and avoid creating too many dummy columns to prevent memory errors.
• Result: Clean, numeric-only dataset ready for regression.

Preprocessing Summary

✅ Preprocessing Done!

Shape after preprocessing: (85907, 3)

Explanation:
Shape = (Rows, Columns)
Rows = number of customer interactions
Columns = number of features (predictors) + target (CSAT Score)

Step 3: Define Features and Target

• Separate features (X) and target (y).
• Features are the predictor variables (e.g., Item_price, connected_handling_time).
• Target is CSAT Score.

Features and Target

Features Shape: (85907, 2) → Predictor variables used for regression

Target Shape: (85907,) → CSAT Score values to predict

Step 4: Train/Test Split

• Split the dataset into:
  • Training set: 80% → to fit the model
  • Testing set: 20% → to evaluate the model
• Use train_test_split() with a random state for reproducibility.

Train/Test Split Completed

Training Samples: 68,725 → Used to train the regression model

Testing Samples: 17,182 → Used to evaluate model predictions

Step 5: Build Linear Regression Model

• Use sklearn.linear_model.LinearRegression to train the model on the training set.
• Linear Regression formula:
y = β0 + β1*x1 + β2*x2 + ⋯ + βn*xn

Where:

• y = predicted target (CSAT Score)
• xi = features
• βi = coefficients (effect of each feature on CSAT Score)
• β0 = intercept

Linear Regression Coefficients

Item_price: -0.000021 → Effect on CSAT Score per unit change in Item_price

connected_handling_time: 0.000140 → Effect on CSAT Score per unit change in connec

Step 6: Predict on Testing Set

• Calculate predicted CSAT scores: y_pred = model.predict(X_test)
• Calculate residuals: residual = y_test - y_pred

Model Evaluation Metrics (Manual Calculation)

Formulas used:

• MAE (Mean Absolute Error) = (1/n) Σ |y_i - ŷ_i|
• MSE (Mean Squared Error) = (1/n) Σ (y_i - ŷ_i)²
• RMSE (Root Mean Squared Error) = √MSE
• Explained Variance = 1 - var(residuals) / var(y_test)

Step 7: Evaluate Model

• MAE (Mean Absolute Error): Average absolute difference between predicted and actual values
  MAE = (1/n) Σ |yi - ŷi|
• MSE (Mean Squared Error): Average squared difference
  MSE = (1/n) Σ (yi - ŷi)^2
• RMSE (Root Mean Squared Error): Square root of MSE
  RMSE = √MSE
• Explained Variance: How much variance of target is explained by model

Model Evaluation Metrics (Values)

MAE: 1.04 → Average absolute prediction error

MSE: 1.87 → Average squared prediction error

RMSE: 1.37 → Standard deviation of prediction errors

Explained Variance: 0.01 → Portion of variance explained by model

Step 8: Visualize Results

• Predicted vs Actual plot: Shows how well predictions match real CSAT Scores.
• Residual plot: Shows the difference between actual and predicted values to check for patterns/errors.

Step 9: Display Predicted vs Actual Table

• Create a DataFrame with Actual, Predicted, Residual columns.
• Show first 10 rows for clarity.

Predicted vs Actual CSAT

📌 Project Milestone – Week 5

"Build your first baseline regression model."

Explanation in context of your project (E-commerce Recommendation System):

• ✅ Completed data preprocessing (cleaning, encoding, and handling missing values).
• ✅ Selected features: Item_price, connected_handling_time and target: CSAT Score.
• ✅ Split the dataset into training (80%) and testing (20%) sets.
• ✅ Trained a Linear Regression model as the first predictive baseline.
• ✅ Evaluated model using MAE, MSE, RMSE, Explained Variance to provide reference performance metrics.

This milestone demonstrates the first working predictive model for your semester-long project and sets the foundation for further analysis and model optimization.