This Project predicts the price of a Real Estate property on the basis of Features like: area_type, location, total_sqft, balcony, bathroom and BHK
https://gaurav-van-house-price-predictor-streamlit-heroku-app-g56zmy.streamlitapp.com/
import numpy as np
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
from scipy import stats
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LinearRegression
from sklearn.tree import DecisionTreeRegressor
from sklearn.ensemble import RandomForestRegressor
from sklearn.metrics import mean_squared_error as mse
from sklearn.model_selection import cross_val_score
from sklearn.model_selection import KFold
%matplotlib inline
#stored within the notebook
import matplotlib
matplotlib.rcParams['figure.figsize'] = (20, 10)
import warnings
warnings.filterwarnings('ignore')Data From Kaggle: https://www.kaggle.com/datasets/amitabhajoy/bengaluru-house-price-data
Data_frame = pd.read_csv("Bengaluru_House_Data.csv", na_values=[' ?'])
Data_frame.head(10)
In predicting Price of a property, 'availability' and 'society' are not THAT important factor in comparison to other features
Data_frame_copy = Data_frame.copy()
Data_frame.drop(['availability', 'society'], axis=1, inplace=True)Data_frame.isnull().sum()
Problem 1: 1, 16, 73 are negligible amount in front of 13320 enteries so we can drop,
in the case of balcony -> 609 we might wanna replace them with either mean or median depending on outliers# Droping Missing Values
Data_frame.dropna(subset = ['location', 'size', 'bath'], inplace=True)
Data_frame.shape
#Replacing Missing Values
sns.boxplot(x='balcony', data=Data_frame)
plt.title("Balcony-BoxPlot")
plt.show()
#Replacing the missing data with mean
Data_frame['balcony'].replace(np.nan, Data_frame['balcony'].mean(), inplace=True)Problem 2: In size feature, some enteries are in the form of 2 BHK and some are in 2 bedrooms ,
Both are same so we can solve this by taking only the numeric value from the size featureData_frame['BHK'] = Data_frame['size'].apply(lambda x: int(x.split(' ')[0]))
Data_frame.drop(['size'], axis=1, inplace=True)
Data_frame.head()Problem 3: total_sqft should be float or numeric but here it is object [ because of dim1-dim2 input ] [and inputs like
sq Yards, Grounds, etc..] The data is Not Structured
Solution: so the best way would be to find an avg value of all dim1-dim2 input and replace it
and convert the different types of meaurement to sqft while doing so, all the values going in the
total_sqft feature will make the data type of that feature -> float-
Creating Function to Convert ranges to a single standard value : Sq.Meter, Perch, Sq.Yards, Acres, Cents, Guntha, Grounds.
1 Sq.Meter = 10.76 sqft,1 Perch = 272.25 sqft,1 Sq.Yard = 9 sqft,1 Acre = 43560.04 sqft,1 Cent = 435.56 sqft,1 Guntha = 1089 sqftand1 Ground = 2400.35 sqft -
Adding a Column / Feature which is important for future use [outlier treatment ] + gives a proper insight of a property
Data_frame['price_per_sqft'] = Data_frame['price']*100000 / Data_frame['total_sqft']
Outliers are those data points those are way off from our main data set [ abnormal data points ] Now they can be of Type1-> Data points / numerical and
Type2 -> Domain error [ abnormality in the Domain Knowledge ]. Type1 and Type2 are similar, Not that Different
-
Fixing Type 1 Outliers: Problem in total_sqft, bathroom, price, BHK and price_per_sqft
# Applying Quantile Based Flooring and capping lower_bound = Data_frame['total_sqft'].quantile(0.10) upper_bound = Data_frame['total_sqft'].quantile(0.90) Data_frame['total_sqft'] = np.where(Data_frame['total_sqft'] < lower_bound, lower_bound, Data_frame['total_sqft']) Data_frame['total_sqft'] = np.where(Data_frame['total_sqft'] > upper_bound, upper_bound, Data_frame['total_sqft']) # Bathroom - small quanitites of Outliers so Replace them with median median = Data_frame['bathroom'].quantile(0.50) upper_out = Data_frame['bathroom'].quantile(0.95) Data_frame['bathroom'] = np.where(Data_frame['bathroom'] > upper_out, median, Data_frame['bathroom']) # Applying Quantile Based Flooring and capping lower_bound = Data_frame['price'].quantile(0.10) upper_bound = Data_frame['price'].quantile(0.90) Data_frame['price'] = np.where(Data_frame['price'] < lower_bound, lower_bound, Data_frame['price']) Data_frame['price'] = np.where(Data_frame['price'] > upper_bound, upper_bound, Data_frame['price'])
-
Fixing Type 2 Outliers
In a General Real Esate Property / House, the number of Bathrooms depends on number of Bedrooms [BHK]. The equations in general is total Bathroom <= BHK + 1 [1 - extra for Guest] It is unusual to have 2 more bathrooms than number of bedrooms in a home
Data_frame = Data_frame[(Data_frame['bathroom'] < (Data_frame['BHK'] + 2))] Data_frame.shape Data_frame['balcony'] = Data_frame['balcony'].astype('int')
Problem here is that in some cases - price of 2bhk is more than price of 3bhk for similar sqft area and same location and hence is a unexpected error or outlier. Here Prices of 2bhk are more than 3bhk for similar or same total_sqft area, it means price_per_sqft of 2bhk should be more than 3bhk for the same location and similar total_sqft area So now we can remove those 3BHK whose price_per_sqft is less than mean price_per_sqft of 2 BHK. We here are dealing with same location [ because Different Location will affect the price ]
def remove_bhk_outliers(df): exclude_indices = np.array([]) for location, location_df in df.groupby('location'): bhk_stats = {} for bhk, bhk_df in location_df.groupby('BHK'): bhk_stats[bhk] = { 'mean': np.mean(bhk_df.price_per_sqft), 'std': np.std(bhk_df.price_per_sqft), 'count': bhk_df.shape[0] } for bhk, bhk_df in location_df.groupby('BHK'): stats = bhk_stats.get(bhk-1) if stats and stats['count']>5: exclude_indices = np.append(exclude_indices, bhk_df[bhk_df.price_per_sqft<(stats['mean'])].index.values) return df.drop(exclude_indices,axis=0) Data_frame = remove_bhk_outliers(Data_frame) Data_frame.shape
In general Square ft per Bedroom is 300 anything less than that is suspicious and can be declared as outlier
Data_frame=Data_frame[~(Data_frame.total_sqft/Data_frame.BHK<300)] Data_frame.shape
num_corr = Data_frame[['total_sqft', 'bathroom', 'balcony', 'BHK']].corr()
sns.heatmap(num_corr, cmap='RdBu')
plt.title("Heatmap of Correlation between numeric Independent Features")
plt.show()
The P value from pearson correlation is 0 ( < 0.05 ) so our Correlation coef are statically significant balcony the weakest while total_sqft the strongest Correlation is not equal to causation -> Balcony might be weakest but is an important feature in deciding the price of a Property
Average price of category from categorical features
df3 = Data_frame[['area_type','location', 'price']]
df_p = df3.groupby(['area_type','location'], as_index=False).mean().sort_values(ascending=False, by='price')
df_p.set_index('area_type', inplace=True)
df_p.rename(columns={'price':'avg-price'}, inplace=True)
df_pivot = pd.pivot_table(df_p, index='area_type', columns='location', values='avg-price')
df_pivot.fillna(0, inplace=True)
df_pivot 
Encoding the Categoricals features - area_type and location
one_dum = pd.get_dummies(Data_frame['area_type'])
Data_frame = pd.concat([Data_frame, one_dum], axis=1)
Data_frame.drop(['area_type'], axis=1, inplace=True)
ne_dum = pd.get_dummies(Data_frame['location'])
Data_frame = pd.concat([Data_frame, ne_dum], axis=1)
Data_frame.drop(['location'], axis=1, inplace=True)As I am dealing with Regression problem, that too linear models so feature scalling is ignored here
LinearModel = LinearRegression()
LinearModel.fit(X_train, Y_train)
Y_pred = LinearModel.predict(X_test)
Y_pred- R^2 Value: 0.824561107732018
- MSE Value: 669.3971397935138
- RMSE Value: 25.87271032
Tree_model = DecisionTreeRegressor(max_depth=5)
Tree_model.fit(X_train, Y_train)
Y_hat = Tree_model.predict(X_test)- R^2 Value: 0.8004133365466866
rf = RandomForestRegressor(n_estimators=105)
rf.fit(X_train, Y_train)
Y_pred_rf = rf.predict(X_test)- R^2 Value: 0.8414810182841019
import lazypredict
from lazypredict.Supervised import LazyRegressor
clf = LazyRegressor(verbose=0,ignore_warnings=True, custom_metric=None)
models_test,predictions_test = clf.fit(X_train, X_test, Y_train, Y_test)import pickle
with open('bangalore_home_prices_model.pickle', 'wb') as obj:
pickle.dump(LinearModel, obj)
import json
columns = {
'Columns': [col.lower() for col in X.columns]
}
with open("Columns.json", 'w') as f:
f.write(json.dumps(columns))with open(
r"C:\Users\Asus\PycharmProjects\Real_Estate_price_prediction\Model\bangalore_home_prices_model.pickle",
'rb') as f:
__model = pickle.load(f)
with open(r"C:\Users\Asus\PycharmProjects\Real_Estate_price_prediction\Model\Columns.json", 'r') as obj:
__data_columns = json.load(obj)["Columns"]
__area_types = __data_columns[4:8]
__locations = __data_columns[8:]def get_predicted_price(area_type, location, sqft, balcony, bathroom, BHK):
try:
area_index = __data_columns.index(area_type.lower())
loc_index = __data_columns.index(location.lower())
except ValueError as e:
area_index = -1
loc_index = -1
lis = np.zeros(len(__data_columns))
lis[0] = sqft
lis[1] = bathroom
lis[2] = balcony
lis[3] = BHK
if loc_index >= 0 and area_index >= 0:
lis[area_index] = 1
lis[loc_index] = 1
price = round(__model.predict([lis])[0], 2)
strp = ' lakhs'
if math.log10(price) >= 2:
price = price / 100
price = round(price, 2)
strp = " crores"
return str(price) + strpdef main():
global result
st.title("Bangalore House Price Predictor")
html_temp = """
<div>
<h2>House Price Prediction ML app</h2>
</div>
"""
st.markdown(html_temp, unsafe_allow_html=True)
total_sqft = st.text_input("Total_sqft")
balcony = st.text_input("Number of Balconies")
bathroom = st.text_input("Number of Bathrooms")
BHK = st.text_input("BHK")
area_type = st.selectbox("Area Type", __area_types)
location = st.selectbox("Location", __locations)
if st.button("Predict"):
result = get_predicted_price(area_type, location, total_sqft, balcony, bathroom, BHK)
st.success(f"Price = {result}")