You can veiw the Jupyter notebook file(.ipynb) here

Analysis of Auto MPG: Predicting High vs. Low Gas Mileage

This notebook explores the Auto.csv dataset to understand the factors that determine whether a car has high or low gas mileage. Our first goal is to transform the continuous mpg (miles per gallon) variable into a binary, categorical variable. Then, we will perform a deep exploratory data analysis (EDA) to see which other features, like weight, horsepower, or origin, are the strongest predictors of this new “high MPG” or “low MPG” status.

1. Data Preparation and Setup

(a) Creating the `mpg01` Binary Variable

First, we import the necessary libraries (Pandas, NumPy, Matplotlib, and Seaborn) and load the Auto.csv dataset.

Before we can analyze the data, we must perform two key preprocessing steps:

Data Cleaning: The horsepower column, which is critical for our analysis, was loaded as a text column because 5 entries were missing and marked with a '?'. We replace these markers with NaN (Not a Number) and then fill these missing spots using the dataset’s median horsepower of 93.5. Using the median makes this step robust to outliers.
Target Variable Creation (Problem 14a): We binarize our target variable, mpg. We first calculate the median mpg for the entire dataset, which is 23.0. We then create a new column called mpg01, which is 1 for cars with mpg > 23.0 (high mileage) and 0 for cars with mpg <= 23.0 (low mileage). This split creates a nearly balanced dataset of 206 ‘low MPG’ cars and 191 ‘high MPG’ cars, which is ideal for prediction.

Finally, we drop the name column, as it’s a unique identifier and not a predictive feature.

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns

# Load the dataset
file_path = 'Auto.csv'
auto_df = pd.read_csv(file_path)

# --- 1. Data Cleaning ---
# Handle 'horsepower'
auto_df['horsepower'] = auto_df['horsepower'].replace('?', np.nan)
auto_df['horsepower'] = pd.to_numeric(auto_df['horsepower'])

# Get median horsepower
median_hp = auto_df['horsepower'].median()

# Impute with median
# FIX for FutureWarning: Use assignment instead of inplace=True on a column
auto_df['horsepower'] = auto_df['horsepower'].fillna(median_hp)

# Drop 'name' column as it's an identifier
auto_df_cleaned = auto_df.drop('name', axis=1)

# --- 2. Part (a): Create mpg01 ---
# Calculate median mpg
mpg_median = auto_df_cleaned['mpg'].median()

# Create 'mpg01' (1 if mpg > median, 0 if mpg <= median)
auto_df_cleaned['mpg01'] = (auto_df_cleaned['mpg'] > mpg_median).astype(int)

# --- 3. Part (b): Graphical Exploration ---
# Set plot theme
sns.set_theme(style="whitegrid")

# --- Plot 1: Continuous Features (Boxplots) ---
continuous_features = ['displacement', 'horsepower', 'weight', 'acceleration']

fig, axes = plt.subplots(2, 2, figsize=(16, 12))
fig.suptitle('Continuous Features vs. MPG Category (0=Low, 1=High)', fontsize=20, y=1.03)
axes = axes.flatten()

for i, feature in enumerate(continuous_features):
    # FIX for FutureWarning: Assign 'x' to 'hue' and set legend=False
    sns.boxplot(ax=axes[i], x='mpg01', y=feature, data=auto_df_cleaned, 
                palette='pastel', hue='mpg01', legend=False)
    
    axes[i].set_title(f'MPG Category vs. {feature.title()}', fontsize=14)
    axes[i].set_xlabel('MPG Category (0=Low, 1=High)', fontsize=12)
    axes[i].set_ylabel(feature.title(), fontsize=12)

plt.tight_layout()
plt.savefig('mpg01_vs_continuous_boxplots.png')
print("Saved boxplots to 'mpg01_vs_continuous_boxplots.png'")

# --- Plot 2: Categorical Features (Count Plots) ---
categorical_features = ['cylinders', 'year', 'origin']

fig, axes = plt.subplots(1, 3, figsize=(20, 6))
fig.suptitle('Categorical Features vs. MPG Category (0=Low, 1=High)', fontsize=20, y=1.05)

for i, feature in enumerate(categorical_features):
    sns.countplot(ax=axes[i], x=feature, hue='mpg01', data=auto_df_cleaned, palette='pastel')
    
    if feature == 'origin':
        # FIX for UserWarning: Set ticks explicitly before setting labels
        axes[i].set_xticks([0, 1, 2]) # Assuming origin values 1, 2, 3 map to indices 0, 1, 2
        axes[i].set_xticklabels(['1: USA', '2: Europe', '3: Japan'])
    
    axes[i].set_title(f'MPG Category by {feature.title()}', fontsize=14)
    axes[i].set_xlabel(feature.title(), fontsize=12)
    axes[i].set_ylabel('Count', fontsize=12)
    axes[i].legend(title='MPG01', labels=['0 (Low)', '1 (High)'])

plt.tight_layout()
plt.savefig('mpg01_vs_categorical_countplots.png')
print("Saved count plots to 'mpg01_vs_categorical_countplots.png'")

# --- Plot 3: Correlation Matrix (Heatmap) ---
# Calculate correlation matrix
corr_matrix = auto_df_cleaned.corr()

plt.figure(figsize=(12, 10))
sns.heatmap(corr_matrix, annot=True, fmt='.2f', cmap='coolwarm', linewidths=0.5, annot_kws={"size": 10})
plt.title('Correlation Matrix of All Features', fontsize=16)
plt.xticks(rotation=45, ha='right')
plt.yticks(rotation=0)
plt.savefig('correlation_heatmap.png')
print("Saved correlation heatmap to 'correlation_heatmap.png'")

Saved boxplots to 'mpg01_vs_continuous_boxplots.png'
Saved count plots to 'mpg01_vs_categorical_countplots.png'
Saved correlation heatmap to 'correlation_heatmap.png'

png

2. Graphical Exploration (Problem 14b)

Now, we investigate the association between our mpg01 variable and the other features. The goal is to identify which features are most predictive of whether a car has high or low gas mileage.

Continuous Features: Boxplots

We use boxplots to compare the distributions of continuous features (displacement, horsepower, weight, acceleration) for high (mpg01=1) and low (mpg01=0) mileage cars.

Findings:

Analysis of Auto MPG: Predicting High vs. Low Gas Mileage

1. Data Preparation and Setup

(a) Creating the `mpg01` Binary Variable

2. Graphical Exploration (Problem 14b)

Continuous Features: Boxplots

Categorical Features: Count Plots

Overall Correlation: Heatmap

Conclusion: Most Useful Features

Analysis of Auto MPG: Predicting High vs. Low Gas Mileage

1. Data Preparation and Setup

(a) Creating the mpg01 Binary Variable

2. Graphical Exploration (Problem 14b)

Continuous Features: Boxplots

Categorical Features: Count Plots

Overall Correlation: Heatmap

Conclusion: Most Useful Features

(a) Creating the `mpg01` Binary Variable