Auto MPG — Fuel Economy Analysis

392 vehicles · Model years 1970–1982 · UCI Machine Learning Repository

Overview
Predictor
Distributions
Correlation
Box Plots
Trends
Regression
Diagnostics
Neural Network
Between 1970 and 1982, the American auto industry faced its biggest upheaval since the Model T. Two oil crises forced carmakers to rethink everything — engine size, weight, country of origin. Average fuel economy nearly doubled in just twelve years. The 392 vehicles in this dataset capture that transformation in numbers.
+81%
MPG improvement, 1970 → 1982
Average rose from 17.7 to 32.0 MPG as regulations tightened and engines shrank
52%
Japanese efficiency advantage
Japanese cars averaged 30.5 MPG vs 20.0 MPG for American-made vehicles
−7.6
MPG lost per 1,000 lbs of weight
The lightest cars (under 2,225 lbs) averaged 32 MPG — the heaviest just 15 MPG
'80
Peak efficiency year
1980 models averaged 33.8 MPG — the highest in the dataset, driven by post-crisis downsizing
Loading…
Descriptive Statistics
Loading…

Average MPG by Country of Origin

MPG Trend by Origin over the Years

How Each Measurement Is Spread
Loading…
How Features Move Together

Correlation Heatmap (Pearson r)

Loading…

Strength of relationship with MPG

Pairwise Scatter Plots
MPG Spread Within Groups

MPG by Number of Cylinders

MPG by Country of Origin

MPG by Model Year — Median with Interquartile Range

Actual vs Predicted MPG

Feature Weights — positive = raises MPG, negative = lowers it

Multicollinearity Check (Variance Inflation Factors)

VIF above 10 means this feature overlaps heavily with others — displacement shares most of its information with weight and cylinders.

Residuals vs Fitted — should be a random cloud around zero

Normal Q-Q Plot — dots on the line means residuals are normally distributed

Residual Distribution — should look like a bell curve

Cook's Distance — red bars are unusually influential data points

Predict MPG for Any Vehicle

Move the sliders to describe a vehicle — the model updates its prediction instantly.

3 cyl8 cyl
68455
46 hp230 hp
1,613 lbs5,140 lbs
8 sec (fast)25 sec (slow)
'70 (dataset start)'95 (extrapolated)
-- MPG
9 MPG55 MPG
How Your Vehicle Compares to the Dataset Average

Feature profile — percentage position within the dataset range

A neural network learns by passing data through layers of connected nodes. Each connection has a weight — a number the network adjusts during training to reduce its prediction error. Unlike linear regression, it can capture curved, non-linear relationships in the data. Configure the architecture below, hit Train, and watch it learn in real time.

Network Architecture

Epoch 0 / 100

Live Architecture Diagram

Training Loss — falling means the network is learning

Neural Net vs Linear Regression vs Actual MPG

Explore How the Data Separates

Pick any two features for the axes. Points are coloured by MPG — red = low, green = high. After training, switch colour mode to see where the neural network's predictions agree or disagree with reality.

Train the network to unlock "NN Predicted MPG" and "NN Error" colour modes.
Predict with the Trained Network

Train the network first, then use these sliders to predict.

Neural Network
--
MPG
Linear Regression
--
MPG
Difference
--
MPG