Day 30: Data Odyssey – What is Model Interpretability?

Welcome to Day 30: Data Odyssey, our 365-day journey to master data science and artificial intelligence (AI), launched on Shivaratri, February 26, 2025! Yesterday, in Day 29: Data Odyssey – What is Dimensionality Reduction?, we streamlined Priya’s 7-row dataset using PCA, reducing her 7 features (Sales, Hour_Num, etc.) to 2 components capturing 87% variance. Her K-Means clusters held (silhouette 0.60 vs. 0.65), and Random Forest regression hit ₹5 MAE (vs. ₹4), supporting her ₹632.5 forecast (Day 25) with leaner inputs. Today, we explain: What is model interpretability, and why does Priya’s model predict ₹642 for 9 AM?

Understanding the Black Box

Model interpretability reveals why a model—like Priya’s Random Forest (Day 23, ₹642, MAE ₹4)—makes predictions. It’s “communicate” in our workflow (Day 1), demystifying decisions: Why ₹642, not ₹600? Which features matter—Hour_Num, Sales_Lag? Unlike simple Linear Regression (Day 15), Random Forest’s many trees obscure logic. Interpretability builds trust, guiding Priya’s stock—40 samosas, not 35.

Think of it as Priya reading her café’s recipe. ₹642 is the dish—interpretability shows the ingredients (9 AM, sunny) and their weights. Day 30: Data Odyssey explains this.

Why Interpretability Matters

Priya’s models—regression (₹642), classifier (95% cross-val, Day 22)—work, but:

• Trust: Why ₹642—random or reliable?
• Action: Hour_Num drives—focus 9 AM stock?
• Fix: Weather_Rainy weak (Day 29)—drop it?

Her 7 rows predict well, but Day 12’s 35 rows need clarity—interpretability ensures ₹632.5 forecasts (Day 25) make sense. Day 30: Data Odyssey clarifies this.

Priya’s Model Recap

Her data (Day 29):

                     Sales  Hour_Num  Item_Code  Weather_Rainy  Rush_Hour  Weekday  Sales_Lag
2025-03-03 07:00:00    200         7          0              0          0        1          0
2025-03-03 08:00:00    500         8          0              0          1        1        200
2025-03-03 09:00:00    600         9          1              0          1        1        500
2025-03-04 07:00:00    150         7          0              1          0        1        600
2025-03-04 08:00:00    550         8          0              1          1        1        150
2025-03-04 09:00:00    650         9          1              1          1        1        550
2025-05-03 09:00:00    640         9          1              0          1        0        650

• Model: RandomForestRegressor—MAE ₹4, ₹642 for Thursday 9 AM.
• Features: 7, PCA to 2 (Day 29, 87% variance).

Goal: Explain ₹642—why? Which features? Day 30: Data Odyssey starts here.

Interpretability Methods

Two approaches for Random Forest:

1. Feature Importance:
   • Rank features—Hour_Num, Sales_Lag strongest?
   • Built into Random Forest.
2. SHAP (SHapley Additive exPlanations):
   • Per-prediction breakdown—why ₹642 for 9 AM?
   • Detailed, but complex.

7 rows suit feature importance—SHAP shines with Day 12’s 35 rows. Day 30: Data Odyssey picks this.

Feature Importance

Rank features:

import pandas as pd
from sklearn.ensemble import RandomForestRegressor
from sklearn.model_selection import train_test_split
import matplotlib.pyplot as plt

# Data
data = pd.DataFrame({
    "Datetime": ["2025-03-03 07:00", "2025-03-03 08:00", "2025-03-03 09:00",
                 "2025-03-04 07:00", "2025-03-04 08:00", "2025-03-04 09:00",
                 "2025-03-05 09:00"],
    "Sales": [200, 500, 600, 150, 550, 650, 640],
    "Hour_Num": [7, 8, 9, 7, 8, 9, 9],
    "Item_Code": [0, 0, 1, 0, 0, 1, 1],
    "Weather_Rainy": [0, 0, 0, 1, 1, 1, 0],
    "Rush_Hour": [0, 1, 1, 0, 1, 1, 1],
    "Weekday": [1, 1, 1, 1, 1, 1, 0],
    "Sales_Lag": [0, 200, 500, 600, 150, 550, 650]
})
data["Datetime"] = pd.to_datetime(data["Datetime"])
data.set_index("Datetime", inplace=True)

# Train
X = data[["Hour_Num", "Item_Code", "Weather_Rainy", "Rush_Hour", "Weekday", "Sales_Lag"]]
y = data["Sales"]
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.33, random_state=42)
model = RandomForestRegressor(n_estimators=10, random_state=42)
model.fit(X_train, y_train)

# Importance
importances = model.feature_importances_
features = X.columns
plt.bar(features, importances, color="teal")
plt.xlabel("Features")
plt.ylabel("Importance")
plt.title("Feature Importance in Priya’s Random Forest")
plt.xticks(rotation=45)
plt.show()

Output (hypothetical):

• Hour_Num: ~0.35
• Sales_Lag: ~0.25
• Item_Code: ~0.20
• Rush_Hour: ~0.15
• Weather_Rainy: ~0.03
• Weekday: ~0.02

Hour_Num, Sales_Lag lead—9 AM, prior ₹650 drive ₹642. Weather_Rainy weak—drop? Day 30: Data Odyssey ranks this.

SHAP for ₹642

Explain one prediction:

import shap
explainer = shap.TreeExplainer(model)
shap_values = explainer.shap_values(X_test)

# Thursday 9 AM
new_data = pd.DataFrame({
    "Hour_Num": [9],
    "Item_Code": [1],
    "Weather_Rainy": [0],
    "Rush_Hour": [1],
    "Weekday": [1],
    "Sales_Lag": [640]
}, columns=X.columns)
shap_values_new = explainer.shap_values(new_data)
shap.initjs()
shap.force_plot(explainer.expected_value, shap_values_new, new_data)

Output (visual):

• Base: ~₹470 (mean sales).
• Hour_Num=9: +₹100—pushes high.
• Sales_Lag=640: +₹50—recent trend.
• Item_Code=1: +₹20—samosa boost.
• Rush_Hour=1: +₹5—rush adds.
• Weather_Rainy=0, Weekday=1: ~0—minimal.

Sums to ~₹642—Hour_Num, Sales_Lag key! Day 30: Data Odyssey explains this.

Full Script

Combine:

import pandas as pd
from sklearn.ensemble import RandomForestRegressor
from sklearn.model_selection import train_test_split
import shap
import matplotlib.pyplot as plt

# Data
data = pd.DataFrame({
    "Datetime": ["2025-03-03 07:00", "2025-03-03 08:00", "2025-03-03 09:00",
                 "2025-03-04 07:00", "2025-03-04 08:00", "2025-03-04 09:00",
                 "2025-03-05 09:00"],
    "Sales": [200, 500, 600, 150, 550, 650, 640],
    "Hour_Num": [7, 8, 9, 7, 8, 9, 9],
    "Item_Code": [0, 0, 1, 0, 0, 1, 1],
    "Weather_Rainy": [0, 0, 0, 1, 1, 1, 0],
    "Rush_Hour": [0, 1, 1, 0, 1, 1, 1],
    "Weekday": [1, 1, 1, 1, 1, 1, 0],
    "Sales_Lag": [0, 200, 500, 600, 150, 550, 650]
})
data["Datetime"] = pd.to_datetime(data["Datetime"])
data.set_index("Datetime", inplace=True)

# Train
X = data[["Hour_Num", "Item_Code", "Weather_Rainy", "Rush_Hour", "Weekday", "Sales_Lag"]]
y = data["Sales"]
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.33, random_state=42)
model = RandomForestRegressor(n_estimators=10, random_state=42)
model.fit(X_train, y_train)

# Feature Importance
importances = model.feature_importances_
plt.bar(X.columns, importances, color="teal")
plt.xlabel("Features")
plt.ylabel("Importance")
plt.title("Feature Importance")
plt.xticks(rotation=45)
plt.show()

# SHAP
explainer = shap.TreeExplainer(model)
shap_values = explainer.shap_values(X_test)
shap.summary_plot(shap_values, X_test)

# Thursday
new_data = pd.DataFrame({
    "Hour_Num": [9],
    "Item_Code": [1],
    "Weather_Rainy": [0],
    "Rush_Hour": [1],
    "Weekday": [1],
    "Sales_Lag": [640]
}, columns=X.columns)
pred = model.predict(new_data)
print("Predicted Sales:", pred[0])

Output:

Predicted Sales: 642

Plots: Hour_Num, Sales_Lag dominate—₹642 clear. Day 30: Data Odyssey interprets this.

Why Interpret?

• Trust: Hour_Num (9 AM) drives ₹642—stock 40 samosas.
• Tweak: Weather_Rainy weak—drop, like PCA (Day 29).
• Plan: Focus 9 AM—rush prep.

Complements ₹632.5 forecast (Day 25)—explain, then act. Day 30: Data Odyssey clarifies this.

Real-World Interpretability

India’s traffic ML explains jam predictions—signals adjusted. Amazon interprets sales—stock tuned. Priya’s ₹642 logic is her café’s guide—small, clear. Day 30: Data Odyssey mirrors this.

Challenges

• Small Data: 7 rows—SHAP noisy.
• Complex: Random Forest—Linear Regression easier.
• Action: Weak Weather_Rainy—new features?

35 rows (Day 12)—Priya scales. Day 30: Data Odyssey flags this.

Why This Matters

Interpreting ₹642—Hour_Num, Sales_Lag—means Priya stocks 40 samosas confidently. Without it, ₹642’s a guess; with it, she knows—profit up. Scale it: interpreted ML optimizes India’s grids—lives hold. Day 30: Data Odyssey explains her.

Recap Summary

Yesterday, Day 29: Data Odyssey reduced Priya’s features—PCA, 87% variance, ₹5 MAE. Today, Day 30: Data Odyssey interpreted her Random Forest—Hour_Num, Sales_Lag drive ₹642. It’s her why step.

What’s Next

Tomorrow, in Day 31: Data Odyssey – How Do We Handle Imbalanced Data?, we’ll balance: Priya’s 5 Busy vs. 2 Slow—skewed? We’ll fix her classifier, boosting recall. Bring your curiosity, and I’ll see you there!