Vehicles Sales Analysis

Comprehensive statistical analysis of vehicle sales data using univariate, bivariate, and multivariate techniques to identify factors influencing sales, resale values, and market segmentation.

🎯 TL;DR

This project performs in-depth statistical analysis on 155+ vehicle models across 30 brands, achieving:

• Best ROI Models: Identified 7 vehicles with superior return on investment (Honda Odyssey, Toyota 4Runner, etc.).
• Key Price Drivers: Price is 70% explained by engine size, horsepower, and vehicle dimensions.
• Market Segmentation: 6 distinct vehicle clusters ranging from economy to luxury segments.
• Sales Leaders: Ford dominates with highest sales volume; Porsche leads in average price ($41.5K).
• Predictive Model: Linear regression model with 70% R² for resale value prediction.

Perfect for automotive analysts, market researchers, and data scientists exploring vehicle market dynamics.

💡 Problem/Motivation

Understanding vehicle sales dynamics is critical for manufacturers, dealers, and consumers, but faces several challenges:

The Market Challenge

• Price Variability: Vehicle prices range from $10K to $100K+ with unclear value drivers.
• Resale Uncertainty: Consumers lack data-driven insights on which vehicles retain value.
• Brand Perception: Is success driven by brand reputation or actual vehicle attributes?.
• Market Segmentation: Difficulty identifying distinct consumer segments and their preferences.
• Investment Decisions: Buyers need quantitative guidance on ROI for vehicle purchases.

The Solution

This analysis provides a data-driven approach to:

• Identify the key attributes that drive vehicle prices and sales volume.
• Quantify resale value patterns and ROI across different vehicle segments.
• Segment the market into distinct clusters with specific characteristics.
• Reveal correlations between vehicle specifications and market performance.
• Provide actionable insights for manufacturers, dealers, and consumers.

Goal: Uncover the relationships between vehicle attributes (engine size, dimensions, horsepower) and market outcomes (sales, price, resale value) using comprehensive statistical methods.

📊 Data Description

Dataset Overview

Source: Kaggle - Car Sales Dataset

Size: 155 vehicle models after cleaning (originally 156)

Coverage: 30 automotive brands with diverse model portfolios

Variables (16 Total)

Categorical Variables (4):

Variable	Type	Description	Categories
Brand	Nominal	Vehicle manufacturer	30 brands (Ford, Toyota, etc.)
Model	Nominal	Specific vehicle model	155 unique models
Type	Binary	Vehicle classification	Passenger (1) / Car (0)
Years_Launch	Discrete	Years since model launch	Derived from Latest_Launch date

Numerical Variables (12):

Variable	Unit	Range	Description
Sales_in_Thousands	Units (000s)	0.1 - 543.8	Total sales volume
Price_in_Thousands	USD (000s)	9.2 - 192.5	Manufacturer's suggested retail price
Resale_Value	USD (000s)	2.3 - 72.0	Resale value after X years
Engine_Size	Liters	1.0 - 8.0	Engine displacement
Horsepower	HP	55 - 450	Engine power output
Wheelbase	Inches	89.8 - 138.7	Distance between front and rear axles
Width	Inches	60.3 - 79.9	Vehicle width
Length	Inches	149.4 - 220.0	Vehicle length
Curb_Weight	Pounds (000s)	1.5 - 5.6	Total vehicle weight
Fuel_Capacity	Gallons	10.3 - 38.0	Fuel tank capacity
Fuel_Efficiency	MPG	15 - 53	Miles per gallon
Power_Performance_Factor	Ratio	23 - 336	Power-to-performance metric

Data Characteristics

• Class Imbalance: 75% Passenger vehicles, 25% Car vehicles.
• Missing Values: 37 missing values handled via mean imputation.
• Outliers: 38 models identified as outliers across various metrics.
• Correlation Patterns: Strong correlations between engine size, horsepower, and price (r > 0.7).

📁 Project Structure

Vehicles-Sales-Analysis/
│
├── Code/
│   └── ccode1.R                    # Univariate & Bivariate Analysis
│   └── code2.R                     # Multivariate Analysis (PCA, FA, Clustering, Regression)
│
├── Data/
│   └── cars.csv                    # Raw dataset (156 observations × 16 variables)
│
├── Documents/
│   └── guidelines.pdf              # Guidelines of the project
│   └── presentation.pdf            # Presentation of the project
│   └── report.pdf                  # Comprehensive analysis report (52 pages)
│
├── requirements.txt                # R package dependencies
└── README.md                       # This file

Key Dependencies

tidyverse==2.0.0
dplyr==1.1.0
ggplot2==3.4.0
corrplot==0.92
e1071==1.7-13
MASS==7.3-60
caret==6.0-94
FactoMineR==2.8
cluster==2.1.4
psych==2.3.3

🔬 Methodology

Analysis Pipeline

Data Cleaning → Univariate Analysis → Bivariate Analysis → Multivariate Analysis → Modeling

Stage 1: Data Cleaning

Preprocessing Steps:

1. Variable Creation: Combined Brand + Model into Brand_Model variable.
2. Type Conversion:
   • Type: "Passenger" → 1, "Car" → 0.
   • Latest_Launch: Converted to Years_Launch (years since launch).
3. Missing Value Handling:
   • Removed row 34 (10 missing values).
   • Imputed remaining NAs with variable means (37 values across 5 variables).
4. Validation: Verified no zero values in numerical variables.

Data Quality Checks:

• ✅ No duplicate records.
• ✅ All categorical variables properly encoded.
• ✅ Numerical ranges validated against domain knowledge.

Stage 2: Univariate Analysis

2.1 Categorical Variables

Brand Distribution:

• Top 5 Brands: Ford (14 models), Dodge (12), Toyota (11), Mercedes-Benz (11), Chevrolet (10).
• Bottom 4 Brands: Infiniti, Jaguar, Saab, Subaru (2 models each).
• Concentration: Top 10 brands hold 53% of all models.

Type Distribution:

• Passenger: 75% (117 models).
• Car: 25% (38 models).

2.2 Numerical Variables

Location Measures:

• Most variables show right-skewed distributions.
• Trimmed means (5%) are lower than regular means, confirming positive outlier influence.
• Example: Sales_in_Thousands → Mean: 48.13, Trimmed Mean: 38.5, Median: 29.0.

Dispersion Measures:

• Highest Variability: Sales_in_Thousands (CV = 128.75%).
• Lowest Variability: Wheelbase (CV = 7.11%).
• Standard Deviations: Range from 5.1 (Engine_Size) to 68.2 (Sales_in_Thousands).

Distribution Characteristics:

• Skewness Analysis:
  • High positive skew: Sales (3.44), Resale_Value (2.28), Price (2.56).
  • Near-symmetric: Width (0.19), Length (0.31), Fuel_Efficiency (-0.04).
• Kurtosis:
  • Heaviest tails: Sales (16.59), Price (9.32), Resale_Value (6.88).
  • Lightest tails: Fuel_Efficiency (3.10), Width (2.54).

Outlier Analysis (38 models identified):

• Sales Outliers (12): Ford F-Series (543.8K units), Ford Explorer, Toyota Camry.
• Price Outliers (9): Mercedes-Benz CL500 ($192.5K), Porsche Carrera Cabrio.
• Engine Size Outliers (3): Dodge Viper (8.0L), Cadillac Escalade (5.7L).
• Horsepower Outliers (2): Dodge Viper (450 HP), Chevrolet Corvette (345 HP).

Stage 3: Bivariate Analysis

3.1 Categorical × Numerical (Type)

T-Test Results (7 variables with p < 0.05):

• Sales_in_Thousands: Cars sell 40% more than Passengers (p < 0.001).
• Engine_Size: Passengers have 15% larger engines (p = 0.003).
• Wheelbase: Passengers are 8% longer (p < 0.001).
• Width: Passengers are 6% wider (p < 0.001).
• Curb_Weight: Passengers are 20% heavier (p < 0.001).
• Fuel_Capacity: Passengers hold 18% more fuel (p < 0.001).
• Fuel_Efficiency: Cars are 12% more efficient (p = 0.002).

3.2 Categorical × Numerical (Brand)

Money Generated (Sales × Price):

• Top 3: Ford ($47.2M), Dodge ($21.8M), Toyota ($18.4M).
• Bottom 3: Porsche ($1.1M), Saab ($1.3M), Jaguar ($1.5M).
• Insight: Success correlates with model volume, not price.

Mean Price by Brand:

• Most Expensive: Porsche ($41.5K), Mercedes-Benz ($38.2K), Lexus ($36.8K).
• Least Expensive: Saturn ($13.2K), Hyundai ($14.5K), Chevrolet ($15.8K).

Brand Age:

• Oldest: Oldsmobile (7.2 years), Mercury (6.8 years), Nissan (6.5 years).
• Newest: Infiniti (2.1 years), Jaguar (2.3 years), Lexus (2.5 years).

3.3 Numerical × Numerical

Return on Investment (ROI) Analysis:

• Created 4-quadrant plot: Price (x-axis) vs Resale_Value (y-axis).
• High ROI Models (7):
  1. Acura CL.
  2. Audi A4.
  3. Honda Odyssey.
  4. Jeep Grand Cherokee.
  5. Mitsubishi 3000GT.
  6. Toyota Avalon.
  7. Toyota 4Runner.
• Interpretation: Low initial price, high resale value relative to segment.

Correlation Matrix (|r| > 0.7):

Resale_Value ←→ Price (r = 0.92)
Resale_Value ←→ Horsepower (r = 0.85)
Resale_Value ←→ Power_Performance_Factor (r = 0.88)

Price ←→ Horsepower (r = 0.84)
Price ←→ Engine_Size (r = 0.73)
Price ←→ Power_Performance_Factor (r = 0.82)

Engine_Size ←→ Horsepower (r = 0.91)
Engine_Size ←→ Width (r = 0.72)
Engine_Size ←→ Curb_Weight (r = 0.78)
Engine_Size ←→ Power_Performance_Factor (r = 0.86)

Horsepower ←→ Power_Performance_Factor (r = 0.98) [REDUNDANT]

Wheelbase ←→ Width (r = 0.76)
Wheelbase ←→ Length (r = 0.87)

Width ←→ Length (r = 0.82)
Width ←→ Curb_Weight (r = 0.84)

Curb_Weight ←→ Fuel_Capacity (r = 0.81)
Curb_Weight ←→ Fuel_Efficiency (r = -0.75)

Fuel_Capacity ←→ Fuel_Efficiency (r = -0.71)

Key Insights:

• Power_Performance_Factor is redundant with Horsepower (r = 0.98).
• Price driven by: Horsepower (r = 0.84), Engine_Size (r = 0.73).
• Larger vehicles (Curb_Weight) have lower fuel efficiency (r = -0.75).

Stage 4: Multivariate Analysis

4.1 Principal Component Analysis (PCA)

Component Selection (Pearson's Criteria: 80% cumulative variance):

• 4 PCs selected explaining 80.4% of total variance.

PC	Variance Explained	Cumulative Variance	Interpretation
PC1	43.2%	43.2%	General vehicle characteristics (size, power, price)
PC2	16.8%	60.0%	Sales vs. Value trade-off
PC3	13.4%	73.4%	Vehicle type differentiation
PC4	7.0%	80.4%	Model age

PC Interpretation:

PC1 (43.2% variance):

• High positive loadings: Engine_Size (0.31), Horsepower (0.32), Width (0.30), Curb_Weight (0.33).
• Negative loading: Fuel_Efficiency (-0.25).
• Meaning: Large, powerful, fuel-inefficient vehicles.

PC2 (16.8% variance):

• High positive loadings: Sales (0.42), Wheelbase (0.38).
• High negative loadings: Resale_Value (-0.45), Price (-0.42).
• Meaning: Popular, affordable vehicles vs. expensive, high-resale luxury cars.

PC3 (13.4% variance):

• High negative loading: Type (-0.76).
• Moderate negative loading: Length (-0.44).
• Meaning: Car vehicles (Type=0) are shorter than Passenger vehicles.

PC4 (7.0% variance):

• High negative loading: Years_Launch (-0.94).
• Meaning: Newer vs. older models.

Brand Contributions:

• PC1 (Vehicle Size/Power): Lincoln (most influential), Infiniti (least influential).
• PC2 (Sales/Value): Porsche (most influential), Dodge/Ford (high influence).
• PC3 (Type): Jeep (most influential, only brand with 0 Passenger vehicles).
• PC4 (Age): Mercury (most influential, oldest models).

4.2 Factor Analysis (FA)

Kaiser-Meyer-Olkin (KMO) Test: 0.83 (excellent suitability for FA).

Three Extraction Methods Compared:

Method 1: Principal Axis (PA):

• PA1 (30% variance): Vehicle dimensions (Wheelbase, Length, Fuel_Capacity).
• PA2 (39% variance): Price/performance (Price, Horsepower, Power_Performance_Factor).
• PA3 (22% variance): Type (negative correlation).
• PA4 (9% variance): Sales.
• RMSR: 0.02 (excellent fit).
• Mean Item Complexity: 1.6 factors per variable.

Method 2: Minimal Residuals (MR):

• MR1 (32% variance): Weight/fuel (Curb_Weight, Fuel_Capacity, Fuel_Efficiency).
• MR2 (38% variance): Price/performance (Resale_Value, Price, Power_Performance_Factor).
• MR3 (5% variance): No high correlations.
• MR4 (25% variance): Dimensions (Wheelbase, Length).
• RMSR: 0.01 (best fit).
• Mean Item Complexity: 1.8 factors per variable.

Method 3: Maximum Likelihood (ML):

• ML1 (38% variance): Price/performance (Resale_Value, Price, Horsepower).
• ML2 (32% variance): Dimensions (Wheelbase, Length).
• ML3 (26% variance): Weight/fuel (Curb_Weight, Fuel_Capacity, Fuel_Efficiency).
• ML4 (5% variance): No high correlations.
• RMSR: 0.02 (excellent fit).
• Mean Item Complexity: 1.8 factors per variable.
• Note: Variables not normally distributed (Shapiro-Wilk p < 0.05), limiting ML reliability.

Best Method: Minimal Residuals (RMSR = 0.01, clearest factor structure).

4.3 Cluster Analysis

4.3.1 K-Means Clustering (Non-Hierarchical):

Optimal K Selection: Elbow method identified K = 6 clusters.

Silhouette Coefficient: 0.54 (reasonably good clustering quality).

Cluster Profiles:

Cluster	Size	Avg Sales	Avg Price	Avg Resale	Key Characteristics
1	52	67.2K	$18.3K	$11.2K	High-volume, affordable, low resale
2	18	22.1K	$32.5K	$21.8K	Low-volume luxury, high resale
3	12	28.4K	$19.7K	$12.5K	Compact economy segment
4	8	35.6K	$47.9K	$28.4K	Premium luxury, largest dimensions
5	21	15.8K	$22.1K	$13.9K	Small, unpopular, specialized
6	44	58.3K	$21.6K	$13.1K	Popular large vehicles, low resale

Cluster Interpretation:

• Cluster 1: Mass-market leaders (Ford, Toyota, Honda).
• Cluster 2: Luxury segment (Mercedes-Benz, Lexus, BMW).
• Cluster 3: Economy compact cars (Saturn, Hyundai).
• Cluster 4: Ultra-luxury (Porsche, high-end Mercedes).
• Cluster 5: Niche/specialty vehicles (low sales).
• Cluster 6: Popular SUVs/trucks (high sales, low resale).

4.3.2 Hierarchical Clustering (Ward's Method):

Silhouette Coefficient: 0.30 (worse than K-means).

Conclusion: K-means provides superior cluster separation for this dataset.

4.4 Linear Discriminant Analysis (LDA)

Assumption Testing:

• Shapiro-Wilk Test: All variables except Length failed normality test (p < 0.05).
• Conclusion: LDA assumptions not met; analysis not performed.

4.5 Linear Regression (Resale Value Prediction)

Model 1: Initial Model (No preprocessing):

Train/Test Split: 80/20.

Results:

• R²: 0.70 (70% variance explained).
• RMSE: 4.6.
• Residual Standard Error: 5.7.
• Significant Predictors: Price_in_Thousands, Curb_Weight, Fuel_Capacity.

Interpretation: Strong predictive power, but high residual variability suggests outlier influence.

Model 2: Standardized Model (Outliers removed):

Preprocessing:

1. Standardized all variables (z-scores).
2. Removed outliers based on Resale_Value boxplot (8 models removed).
3. Verified balanced skewness (~0).

Results:

• R²: 0.54 (54% variance explained, -16% vs initial).
• RMSE: Lower (standardized scale).
• Residual Standard Error: Lower (tighter fit).
• Significant Predictors: Engine_Size, Width, Curb_Weight.

Interpretation: Outlier removal reduced explanatory power but improved prediction accuracy on typical vehicles. Trade-off between capturing market extremes vs. modeling average behavior.

📈 Results/Interpretation

Key Findings Summary

1. Sales Drivers

• Volume Leaders: Ford (47.2M revenue), Dodge (21.8M), Toyota (18.4M).
• Sales Strategy: Success driven by model portfolio size, not premium pricing.
• Type Effect: Cars outsell Passengers by 40% on average.

2. Price Determinants

• Primary Factors (70% of price variance):
  • Horsepower (r = 0.84).
  • Engine_Size (r = 0.73).
  • Power_Performance_Factor (r = 0.82).
• Premium Brands: Porsche ($41.5K avg), Mercedes-Benz ($38.2K), Lexus ($36.8K).

3. Resale Value Patterns

• High Resale Brands: Porsche (72% retention), Mercedes-Benz (68%), Lexus (65%).
• ROI Champions (7 models): Honda Odyssey, Toyota 4Runner, Acura CL.
• Predictors: Price (r = 0.92), Horsepower (r = 0.85), Power_Performance_Factor (r = 0.88).

4. Market Segmentation

6 Distinct Clusters:

1. Mass-Market (33%): High sales, affordable, Ford/Toyota dominance.
2. Luxury (12%): Low volume, high price/resale, Mercedes/Lexus.
3. Economy (8%): Compact, budget-friendly, Saturn/Hyundai.
4. Ultra-Luxury (5%): Highest price/dimensions, Porsche/Mercedes.
5. Niche (14%): Specialized, low sales.
6. Popular Large (28%): SUVs/trucks, high sales but low resale.

5. Correlation Insights

• Redundant Variable: Power_Performance_Factor (r = 0.98 with Horsepower).
• Size Correlation: Wheelbase ↔ Width ↔ Length (r > 0.76).
• Efficiency Trade-off: Curb_Weight ↔ Fuel_Efficiency (r = -0.75).

6. Outlier Models (38 identified)

• Sales: Ford F-Series (543.8K units, 10× median).
• Price: Mercedes-Benz CL500 ($192.5K, 10× median).
• Performance: Dodge Viper (8.0L engine, 450 HP).

💼 Business Impact

For Automotive Manufacturers

Product Development:

• Insight: 70% of price is determined by Engine_Size, Horsepower, and vehicle dimensions.
• Action: Optimize engine specs and sizing to hit target price points.
• ROI: Improve profit margins by 5-10% through data-driven feature prioritization.

Portfolio Strategy:

• Insight: Model volume drives revenue more than premium pricing (Ford vs Porsche).
• Action: Expand model lineup in high-volume segments (Clusters 1 & 6).
• ROI: Increase market share by 15-20% in underserved segments.

Resale Value Engineering:

• Insight: Resale value strongly correlated with initial price and horsepower.
• Action: Design for resale value retention to justify premium pricing.
• ROI: 10-15% higher transaction prices due to lower total cost of ownership.

For Automotive Dealers

Inventory Optimization:

• Insight: 6 distinct market clusters with different sales velocities.
• Action: Stock inventory matching local cluster preferences.
• ROI: 20-30% reduction in days-on-lot for inventory.

Pricing Strategy:

• Insight: 7 models identified with high ROI (low price, high resale).
• Action: Promote ROI champions in marketing materials.
• ROI: 5-10% increase in conversion rates for targeted models.

Trade-In Valuation:

• Insight: Linear regression model predicts resale value with 70% accuracy.
• Action: Use model for data-driven trade-in offers.
• ROI: Improve trade-in profitability by 8-12%.

For Consumers

Purchase Decisions:

• Insight: ROI analysis identifies 7 best-value models.
• Benefit: Save $3K-$8K over 5-year ownership period.
• Tool: Use cluster profiles to match needs with optimal segment.

Resale Planning:

• Insight: Porsche, Mercedes, Lexus retain 65-72% of value.
• Benefit: Reduce depreciation costs by choosing high-retention brands.
• Tool: Predict future resale value within ±$4.6K (RMSE).

For Market Researchers

Segmentation Framework:

• Insight: 6-cluster model with 0.54 silhouette coefficient.
• Application: Reproducible framework for market studies.
• Validation: Aligns with industry segments (economy, luxury, etc.).

Predictive Modeling:

• Insight: 70% R² for resale value prediction.
• Application: Baseline for advanced ML models (ensemble, neural networks).
• Extension: Incorporate temporal data for time-series forecasting.

🚀 Getting Started

Installation

# Clone the repository
git clone https://github.com/pedroalexleite/Vehicles-Sales-Analysis.git
cd Vehicles-Sales-Analysis

# Install R packages
Rscript -e "install.packages(c('tidyverse', 'dplyr', 'ggplot2', 'ggrepel', 'forcats', 'scales', 'gridExtra', 'corrplot', 'e1071', 'MASS', 'caret', 'tidyr', 'FactoMineR', 'cluster', 'dendextend', 'caTools', 'psych', 'car'))"

Running the Analysis

Option 1: Full Analysis

# Open RStudio
# Run code1.R for univariate and bivariate analysis
source("code1.R")

# Run code2.R for multivariate analysis
source("code2.R")

# Total runtime: ~10-15 minutes

Option 2: Specific Analyses

# Univariate analysis only
source("code1.R")  # Stop after line 300

# Correlation matrix only
source("code1.R")  # Jump to line 500

# PCA only
source("code2.R")  # Stop after line 250

# Clustering only
source("code2.R")  # Lines 400-600

Expected Outputs

Visualizations:

• 30+ plots (histograms, boxplots, scatterplots, dendrograms).
• Correlation matrices (heatmaps).
• PCA biplots.
• Cluster visualizations.

Statistical Tables:

• Descriptive statistics (mean, median, SD, CV, skewness, kurtosis).
• T-test results.
• Correlation coefficients.
• PCA loadings and contributions.
• Cluster profiles.
• Regression coefficients.

Models:

• PCA components (4 retained).
• Factor analysis loadings (3 methods).
• K-means clusters (6 clusters).
• Linear regression model (Resale_Value prediction).

🤝 Contributing

Contributions are welcome:

How to Contribute:

1. Fork the repository.
2. Create a feature branch (git checkout -b feature/TimeSeriesAnalysis).
3. Commit your changes (git commit -m 'Add time-series forecasting').
4. Push to the branch (git push origin feature/TimeSeriesAnalysis).
5. Open a Pull Request.