House_Price_Prediction

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

#Read from the directory where you stored the data

data  = pd.read_csv('/kaggle/input/california-housing-prices/housing.csv')

data

#Test to see if there arent any null values
data.info()

#Trying to draw the same number of null values
data.dropna(inplace = True)

data.info()

#From our data, we are going to train and test our data

from sklearn.model_selection import train_test_split

X = data.drop(['median_house_value'], axis = 1)
y = data['median_house_value']

y

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.2)

#Examining correlation between x and y training data
train_data = X_train.join(y_train)

train_data

#Visualizing the above
train_data.hist(figsize=(15, 8))

#Encoding non-numeric columns to see if they are useful and categorical for analysis

train_data_encoded = pd.get_dummies(train_data, drop_first=True)
correlation_matrix = train_data_encoded.corr()
print(correlation_matrix)

train_data_encoded.corr()

plt.figure(figsize=(15,8))
sns.heatmap(train_data_encoded.corr(), annot=True, cmap = "inferno")

train_data['total_rooms'] = np.log(train_data['total_rooms'] + 1)
train_data['total_bedrooms'] = np.log(train_data['total_bedrooms'] +1)
train_data['population'] = np.log(train_data['population'] + 1)
train_data['households'] = np.log(train_data['households'] + 1)

train_data.hist(figsize=(15, 8))

#convert ocean_proximity factors into binary's using one_hot_encoding
train_data.ocean_proximity.value_counts()

#For each feature of the above we will then create its binary(0 or 1)
pd.get_dummies(train_data.ocean_proximity)

#Dropping afterwards the proximity
train_data = train_data.join(pd.get_dummies(train_data.ocean_proximity)).drop(['ocean_proximity'], axis=1)

train_data

#recheck for correlation
plt.figure(figsize=(18, 8))
sns.heatmap(train_data.corr(), annot=True, cmap ='twilight')

#visualize the coordinates
plt.figure(figsize=(15, 8))
sns.scatterplot(x='latitude', 
                y = 'longitude',
                data= train_data, 
                hue='median_house_value', palette='Spectral')

#Combine new features with the ones we already have(using feature engineering)

train_data['bedroom_ratio'] = train_data['total_bedrooms']/train_data['total_rooms']
train_data['household_rooms'] = train_data['total_rooms']/train_data['households']

#show correlation
plt.figure(figsize=(18, 8))
sns.heatmap(train_data.corr(), annot=True, cmap ='ocean')

#train data using linear regression
from sklearn.linear_model import LinearRegression

X_train, y_train = train_data.drop(['median_house_value'], axis=1), train_data['median_house_value']

reg = LinearRegression()

reg.fit(X_train, y_train)

# Assuming 'data' is the original dataset with 'ocean_proximity'
test_data = X_test.join(y_test)
test_data['ocean_proximity'] = data.loc[test_data.index, 'ocean_proximity']

#Join X_test and y_test to form test_data
test_data = X_test.join.join(y_test)

#Apply log transformations
test_data['total_rooms'] = np.log(test_data['total_rooms'] + 1)
test_data['total_bedrooms'] = np.log(test_data['total_bedrooms'] + 1
test_data['population'] = np.log(test_data['population'] + 1)
test_data['households'] = np.log(test_data['households'] + 1)

#One-hot encode 'ocean_proximity' with the same prefix as in training data
#Assuming the training data used no prefix or a different one, adjust accordingly
#test_data = test_data.join(pd.get_dummies(test_data['ocean_proximity'], prefix='')).drop(['ocean_proximity'], axis=1) #prefix =''for no prefix
# or rename columns to match training data after one-hot encoding
test_data = test_data.rename(columns = {
       'ocean_<1H OCEAN': '<1H OCEAN',
       'ocean_INLAND  : 'INLAND',
       'ocean_ISLAND : 'ISLAND',
       'ocean_NEAR BAY: 'NEAR BAY',
       'ocean_NEAR OCEAN : 'NEAR OCEAN'
})

#Create new feature columns
test_data['bedroom_ratio'] = test_data['total_bedrooms']/test_data['total_rooms']
test_data['household_rooms'] = test_data['total_rooms']/test_data['household_rooms]

#test_data = test_data.join(pd.get_dummies(test_data['ocean_proximity']. prefix='')).drop(['ocean_proximity'], axis=1) #prefix='' for no prefix #or rename columns to match training data after one-hot encoding:
test_data = test_data.rename(columns = {
       'ocean_<1H OCEAN': '<1H OCEAN',
       'ocean_INLAND  : 'INLAND',
       'ocean_ISLAND : 'ISLAND',
       'ocean_NEAR BAY: 'NEAR BAY',
       'ocean_NEAR OCEAN : 'NEAR OCEAN'
})

X_test, y_test = test_data.drop(['median_house_value'], axis=1), test_data['median_house_value']

X_test_s = scaler.transfrom(X_test)

reg.score(X_test_s, y_test)

from sklearn.ensemble import RandomForestRegressor
from sklearn.metrics import mean_squared_error

forest = RandomForestRegressor()

forest.fit(X_train_s, y_train)

forest.score(X_test_s, y_test)

from sklearn.model_selection import GridSearchCV

param_grid ={
     'n_estimators': [3, 10, 30],
     'max_features' : [2, 4, 6, 8]

}

grid_search = GridSearchCV(forest, Param_grid, 
                           cv=5,
                           scoring ="neg_mean_squared_error",
                          return_train_score=True)
grid_search.fit(X_train_s, y_train)

grid_search.best_estimator_

grid_search.best_estimator_.score(X_test_s, y_test)