#Importing and configuring CSV.
#Dataset from Kaggle: https://www.kaggle.com/datasets/austinreese/craigslist-carstrucks-data?resource=download
import pandas as pd
import matplotlib.pyplot as plt
pd.set_option('display.max_columns', None)

df = pd.read_csv('vehicles.csv')
df

#Selecting only relevant features for this study
df = df[['price', 'year', 'odometer', 'manufacturer','condition', 'title_status', 'type', 'state']]

df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 426880 entries, 0 to 426879
Data columns (total 8 columns):
 #   Column        Non-Null Count   Dtype  
---  ------        --------------   -----  
 0   price         426880 non-null  int64  
 1   year          425675 non-null  float64
 2   odometer      422480 non-null  float64
 3   manufacturer  409234 non-null  object 
 4   condition     252776 non-null  object 
 5   title_status  418638 non-null  object 
 6   type          334022 non-null  object 
 7   state         426880 non-null  object 
dtypes: float64(2), int64(1), object(5)
memory usage: 26.1+ MB

#Lots of missing data in categories. We'll encode "missing" into the empty values later on in the project and see how the model performs without the data.
#Data will now be visualized to see outliers.

#Price
df['price'].plot(kind='hist', bins=50)
plt.xlabel('Price')
plt.ylabel('Count')
plt.title('Distribution of Price')
plt.show()

Saved output

#Looks like we have some junk prices that need to be filtered out, we'll use quantile to filter out the high end and will make the floor $1,000 to remove the free entries.
prices = df[(df['price'] > 1000) & (df['price'] <= df['price'].quantile(0.99))]
prices['price'].plot(kind='hist', bins=50)
plt.xlabel('Price')
plt.ylabel('Count')
plt.title('Distribution of Price')
plt.show()

Saved output

#Looks better, now the dataset will be updated. 

df = df[(df['price'] > 1000) & (df['price'] <= df['price'].quantile(0.99))]

df['year'].plot(kind='hist', bins=50)
plt.xlabel('Year')
plt.ylabel('Count')
plt.title('Distribution of Vehicle Year')
plt.show()

Saved output

#A few outliers, but this could include classic cars.

#Mileage
df['odometer'].plot(kind='hist', bins=50)
plt.xlabel('Miles')
plt.ylabel('Count')
plt.title('Distribution of Vehicle Mileage')
plt.show()

Saved output

#More outliers.
quantile = df['odometer'].quantile(0.99)
miles = df[(df['odometer'] >= (0)) & (df['odometer'] <= (quantile))]
miles['odometer'].plot(kind='hist', bins=50)
plt.xlabel('Miles')
plt.ylabel('Count')
plt.title('Distribution of Vehicle Mileage')
plt.show()

Saved output

df = df[(df['odometer'] >= (0)) & (df['odometer'] <= (quantile))]

#Lots of missing values were spotted initially, let's visualize.
missing = df['manufacturer'].fillna('MISSING').value_counts()

missing.sort_values().plot(kind='barh')
plt.xlabel('Number of listings')
plt.ylabel('Manufacturer')
plt.title('Listings by manufacturer')
plt.tight_layout()
plt.show()

Saved output

#These values will be one-hot encoded later to have the unknown values be ignored when training.

#Fixing missing values
features = ['manufacturer', 'condition', 'title_status', 'type', 'state']

df[features] = df[features].fillna('unknown')

#Removing Duplicates
df = df.drop_duplicates()

#One Hot Encoding (Condition)
encoded_condition = pd.get_dummies(df['condition'], prefix='Condition', dtype=int)
df = df.drop('condition', axis=1)
df = pd.concat([df, encoded_condition], axis=1)
df

#One Hot Encoding (Title)
encoded_title_status = pd.get_dummies(df['title_status'], prefix='Title Status', dtype=int)
df = df.drop('title_status', axis=1)
df = pd.concat([df, encoded_title_status], axis=1)
df

#One Hot Encoding (Vehicle Type)
encoded_type = pd.get_dummies(df['type'], prefix='Vehicle Type', dtype=int)
df = df.drop('type', axis=1)
df = pd.concat([df, encoded_type], axis=1)
df

#One Hot Encoding (State)
encoded_state = pd.get_dummies(df['state'], prefix='Vehicle State', dtype=int)
df = df.drop('state', axis=1)
df = pd.concat([df, encoded_state], axis=1)
df

#One Hot Encoding (Manufacturer)
encoded_manufacturer = pd.get_dummies(df['manufacturer'], prefix='Vehicle Manufacturer', dtype=int)
df = df.drop('manufacturer', axis=1)
df = pd.concat([df, encoded_manufacturer], axis=1)
df

#Model Training

# Create X and y Dataframes for SKL
X= df.drop('price', axis=1)
y= df['price']

import sklearn
from sklearn.ensemble import RandomForestRegressor
from sklearn.model_selection import train_test_split

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42) #Test size 20% of data

# Random Forrest Classifier
model = RandomForestRegressor(n_estimators=100, random_state=42, n_jobs=-1)
model.fit(X_train, y_train)
y_pred = model.predict(X_test)

import numpy as np
import seaborn as sns
# Sample of 5,000 points to maintain readability
sample = np.random.RandomState(42).choice(len(y_test), 5000, replace=False)

plt.figure(figsize=(8, 6))
sns.scatterplot(x=y_test.iloc[sample], y=y_pred[sample], alpha=0.3, s=10)
plt.plot([0, y_test.max()], [0, y_test.max()], 'r--', label='Regression Line')
plt.xlabel('Actual Price ($)')
plt.ylabel('Predicted Price ($)')
plt.title('Predicted vs Actual Price')
plt.legend()
plt.show()

Saved output

from sklearn.metrics import mean_absolute_error, r2_score
print('Mean Absolute Error:', mean_absolute_error(y_test, y_pred))
print('R^2:', r2_score(y_test, y_pred))
print('Baseline MAE:', mean_absolute_error(y_test, [y_train.median()] * len(y_test)))

Mean Absolute Error: 3656.616633110759
R^2: 0.7834092710869567
Baseline MAE: 10145.964302942595

import matplotlib.pyplot as plt

# Get feature importances from the trained model
importances = pd.Series(model.feature_importances_, index=X_train.columns)

# Combine one-hot columns back into their original feature since original graph was showing all features making it unreadable.
grouped = importances.groupby(lambda col: col.split('_')[0]).sum()
grouped = grouped.sort_values()

# Plot the feature importances
plt.figure(figsize=(10, 6))
plt.barh(grouped.index, grouped.values)
plt.xlabel('Feature Importance')
plt.ylabel('Feature')
plt.title('Feature Importances in Random Forest')
plt.show()

Saved output