Highlights in Science, Engineering and Technology TPCEE 2022

Volume 38 (2023)

House Rent Analysis with Linear Regression Model—— A

Case Study of Six Cities in India
Zonglin Cai1, *, †, Yiqing Zhao2, †
1 United World College of South East Asia, Singapore
2 University of California, Davis, California, USA
* Corresponding Author Email: cai122427@[Link]
†These authors contributed equally.
Abstract. House rent in India has been rising significantly ever since the pandemic hit the country.
In situations like this, it is important for consumers to have the concept of a reasonable rent price,
as otherwise, they may suffer from landlords raising rents deliberately. To resolve this issue, a
prediction model for rent prices is necessary. This study analyses rent data from six cities (Kolkata,
Mumbai, Bangalore, Delhi, Chennai and Hyderabad) in India with multiple variables, including size,
furnishing status, and the number of bathrooms, bedrooms, halls, and kitchens and creates a
prediction model based on the data. The main analytical methods used are linear regression and
logarithmic transformation. This study also includes a general factor analysis based on the data. The
results suggest that this model is reasonably accurate for reference uses, but needs further
improvements if it is to be used commercially.
Keywords: House rent; India; Linear regression model; Factor analysis.

1. Introduction
1.1. Background
Information asymmetry always exists in rental markets, and it has been a problem, especially for
tenants as they can get abused without even knowing. This has been a problem in India particularly.
In fact, the Rent Control Act established by the Indian government was set out to solve this problem
primarily. However, legislation cannot solve the problem entirely because individuals might still be
subjected to smaller information asymmetries. Moreover, some landlords might be struggling to rent
out their house, as they lack the information of market price and set their rent too high. For these
reasons, it is important to have such a model predicting rent prices.
1.2. Related research
So far, there have been plenty of studies on rent prediction using various methods. In 2003,
Dokmeci et al. analyzed rent data in Istanbul, Turkey using hedonic regression [1]. Their study
suggests that external factors like size, number of rooms, bathrooms, balconies, and type of housing
unit do have an impact on the rent price. In more recent years, more sophisticated algorithms are used.
Ming et al. analyzed the rent price in Chengdu, China with XGBoost [2]. Similarly, Li modeled the
rent price in China using the Light Gradient Boosting Model, which is a more efficient variation of
the XGBoost model [3]. Ma et al. expanded on their study by additionally using Random Forest
Regressor and Extra Tree Regressor [4].
Traditional linear regression models are also frequently used in recent studies. In 2019, Kumar
investigated a case study in Ames, Iowa, the United States with linear regression and ridge regression
[5]. In 2020, Tomal modeled the rent prices in Cracow, Poland using OLS regression [6]. He
especially used SAR and GWR-SAR models to determine the spatial heterogeneity of the parameters
and autocorrelation of housing rent. The results showed that the price of rent had a close relationship
with houses’ structure, location and neighborhood.
Moreover, Andrew Coleman and Grant M. Scobie created a simple model of housing rental and
ownership with policy simulations and took New Zealand as an example in 2009 [7]. Their studies
576
Highlights in Science, Engineering and Technology TPCEE 2022
Volume 38 (2023)

indicated that the demand for rental property is much more elastic than the housing demand in total
because of the substitution theory in economics. Therefore, the factor of owner occupancy rate cannot
determine the measurement of the housing market. Another article about the factors that influenced
the price of rents in Beijing, China indicated that the two most important factors were house area and
location [8]. Within the same city, the price of rent was still very different, mainly affected by whether
the house is located in the commercial center of the city. Christina stated that bathroom and bedroom
numbers, whether pets were allowed in the house, square footage, parking area, whether there were
washers or dryers, location and lawn were the eight factors that affect rent prices [9]. Finally, Bogdan
researched the renting landscape in thirty countries around the world in 2018 [10]. He found out that
the number of renters was growing in most of the thirty countries and the prices also kept increasing.
1.3. Objective
Despite a number of studies surrounding the subject of rent forecast and general analysis of factors
that influence house rent, there have been few studies on India’s rental market, which was affected
heavily by the COVID-19 pandemic. Therefore, our study tries to provide a model for rent prediction
in India with linear regression and logarithmic transformation. In addition, we want to have a general
analysis of factors that have a big impact on housing rents in India.

2. Data
Our data is collected from Kaggle. In total, there are 4746 data entries of various house types in
India. The data is collected from 6 cities in India, which are Kolkata, Mumbai, Bangalore, Delhi,
Chennai and Hyderabad. All data is posted on [Link] and is up to date. Most
of them are posted in April, May, June and July of 2022. Table 1 shows the names of the variables
and their corresponding meaning. For Furnishing Status, we used dummy variables for convenience
during data processing.
Table 1. Names and descriptions of each variable
Variable name Description
BHK Number of bedrooms, hall and kitchen
Rent Rent of the houses/apartments/flats in Indian rupees
Size Size of the houses/apartments/flats in square feet
Houses/apartments/flats situated on which floor
Floor
(Ground floor is 0)
City City where the houses/apartments/flats are located.
Furnishing Furnishing status of the houses/apartments/flats (2 means furnished, 1 is semi-
Status furnished, and 0 is unfurnished)
Bathroom Number of bathrooms

3. Methodology
3.1. Linear regression
Multiple linear regression is a linear approach for modeling the relationship between a scalar
response and one or more explanatory variables (also known as dependent and independent variables).
It can be expressed by the following equation (1)
𝑌 =𝑏•𝑋+𝑒 (1)
where Y is the dependent variable, X is the independent variable, b is an unknown parameter and
e is the error term.

577
Highlights in Science, Engineering and Technology TPCEE 2022
Volume 38 (2023)

3.2. Linear regression with logarithmic transformation

Logarithmic transformation in linear regression is usually used when the relationship between
dependent and independent variables is not linear. It can be expressed by the following equation (2),
where Y is the dependent variable, X is the independent variable, a and b are unknown parameters
and e is the error term.
log 𝑌 = 𝑎 + 𝑏 • 𝑋 + 𝑒 (2)
The benefit of this method is that the linear relationship can still be preserved while a non-linear
relationship is actually being handled. This transformation will increase the precision of the model
compared to the original linear regression. Moreover, this transformation can turn highly skewed data
into a more normal distribution, which makes analysis more convenient. This can be shown in Fig. 1
and Fig. 2, which are the distributions of house rent in Bangalore with and without the transformation,
respectively.
More specifically, the transformation we are using is a log-linear model, which is to take the log
of the dependent variable while remaining independent variables unchanged, as we found out that it
is the best way to deal with dummy variables (compared to linear-log and log-log models).

Fig. 1 Distribution of house rent in Bangalore without logarithmic transformation (Photo credit:
Original)

578
Highlights in Science, Engineering and Technology TPCEE 2022
Volume 38 (2023)

Fig. 2 Distribution of house rent in Bangalore with logarithmic transformation (Photo credit:
Original)

4. Results
Our results can be divided into three parts: general analysis on single factors, linear regression
results on different cities, and results of regression with logarithmic transformation. By isolating
individual factors and performing linear regressions between each of them and the rent price, we want
to see what factors are more relevant to rent prices, and then we provide a simplistic model using
linear regression only. At last, we use logarithmic transformation to improve the accuracy of our
model.
4.1. General analysis of single factors
In this study, we first used regression models to analyze the five factors and compared their impact
on housing rent. The following table includes the R values and standard errors of each of the five
regressions.
Table 2. R, R2 and standard error values of 5 variables
BHK Size Floor Furnishing Bathroom
Multiple R 0.369717574 0.413550758 0.215498864 0.146251109 0.441215229
R2 0.136691085 0.17102423 0.04643976 0.021389387 0.194670878
Adjusted R2 0.136509106 0.170849488 0.046238757 0.021183103 0.194501121
Standard Error 72579.76777 71121.90486 76279.27004 77274.71679 70100.18488
Observation 4746 4746 4746 4746 4746

579
Highlights in Science, Engineering and Technology TPCEE 2022
Volume 38 (2023)

Among the 5 factors, the biggest value of R2 is from the bathroom number (0.19), which means
that it has the largest impact on housing rent. The overall order of R2 is bathroom (0.19) > size (0.17) >
BHK (0.14) > floor level (0.05) > furnishing status (0.02). The greater R2 is, the larger of an impact
each variable has on rent price. Interestingly, our results show that whether a house has been furnished
does not influence the house rent by much.
4.2. Effects of different cities
When it comes to another very important factor --- location, we also analyzed R and R2 values in
different cities. In the first part, we compare all the data of the same factor in different cities with
their corresponding house prices to get R and R2, while in the second part, we aggregated the data for
all the different factors in the same city with their housing rents and compared the R2 between
different cities. The number of observations in six cities are 972, 524, 886, 868, 605 and 891
respectively, and the R2 values are 0.75, 0.30, 0.07, 0.58, 0.50 and 0.36, as shown in Table 3.
Table 3. R, R2, standard error values and number of observations of the six cities
Regression Mumbai Kolkata Bangalore Hyderabad Delhi Chennai
Multiple R 0.866266182 0.54864719 0.267578933 0.761630229 0.703898743 0.599804339
R2 0.750417098 0.301013739 0.071598485 0.580080605 0.49547344 0.359765246
Adjusted R2 0.749125261 0.294266767 0.066323477 0.577644878 0.491262033 0.3561481
Standard Error 51352.16277 9356.378713 116006.5973 17180.58125 31056.74366 26535.4225
Observation 972 524 886 868 605 891

From Fig. 3, it is very obvious that Mumbai has the largest R2 value (0.75) followed by Hyderabad
(0.58) while Bangalore has the smallest (0.07). More specifically, it indicates that 75% of the variation
of all dependent variables in Mumbai can be explained by the regression model, however, only 7%
of the variation in Bangalore has a close correlation.

R2
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
Mumbai Kolkata Bangalore Hyderabad Delhi Chennai

Fig. 3 Comparison of R2 values between the six cities (Photo credit: Original)
4.3. Logarithmic transformation
After performing another round of regression with logarithmic transformation, we have some
additional results.
Table 4 shows the coefficients of the regression for each of the cities. From these coefficients, we
can derive a model for each city. For example, the model for Kolkata can be expressed as
log 𝑟𝑒𝑛𝑡 = 0.0792783 ∗ BHK + 0.00021176 ∗ Size + (−0.0635775) ∗ Floor + 0.03635651
∗ Furnishing + 0.12363036 ∗ Bathrooms + 3.4873161

580
Highlights in Science, Engineering and Technology TPCEE 2022
Volume 38 (2023)

Table 4. Coefficients of the variables of each city

Kolkata Mumbai Bangalore Delhi Chennai Hyderabad
Intercept 3.4873161 3.7115403 4.13178995 3.6699341 3.65330958 3.69299258
BHK 0.0792783 0.06357035 0.09007733 0.07253449 0.08497004 0.08972869
Size 0.00021176 0.00025701 0.00032202 0.00012619 0.00025124 0.00018509
Floor -0.0635775 -0.1129652 -0.1409431 -0.0483275 -0.0594339 -0.0574164
Furnishing 0.03635651 0.0400852 0.07481495 0.07248787 0.06042515 0.05895562
Bathrooms 0.12363036 0.07367975 0.03918431 0.1749855 0.04648302 0.03366166

Except for R and R2, we also used p-value to determine the correlation between independent
variables and dependent variables.
Table 5 shows the p-values of variables for each of the 6 cities. As we can see, all the p-values are
below 0.01, meaning that all observations are statistically significant, and there are strong
relationships between independent and dependent variables. Interestingly, the p-value of the variable
Bathrooms in Bangalore, Chennai and Hyderabad are significantly larger than the others, all
exceeding 0.001, indicating that the relationship between Bathrooms and the rent price might not be
so strong. Moreover, the p-value of Floor in Delhi is also greater than 0.001, which means that floor
might not be as relevant to the rent price. Lastly, p-values of Size in all six cities are considerably
smaller than other variables, showing that among all variables Size has the strongest relationship to
rent, which makes sense intuitively as well.
Table 5. P-values of the variables of each city
P-value
Kolkata Mumbai Bangalore Delhi Chennai Hyderabad
Intercept 0 0 0 0 0 0
BHK 1.8727E-08 6.8572E-06 8.3828E-09 7.6176E-06 3.7144E-08 2.1879E-13
Size 4.4169E-16 4.6736E-51 3.7473E-40 1.0881E-13 7.6665E-47 1.1809E-42
Floor 6.1915E-05 4.5656E-22 7.4858E-14 0.00432628 2.4392E-06 9.9201E-06
Furnishing 0.00141311 4.2049E-05 5.9122E-18 1.9134E-09 4.9372E-10 2.9086E-11
Bathrooms 3.3772E-12 3.3935E-07 0.00361025 2.5992E-23 0.00216301 0.00764831

Table 6 shows the adjusted R2 value of rent in the six cities with and without logarithmic
transformation. We can clearly see that other than Mumbai, which suffered a minor decrease in R2
value, all other five models had considerable increases to their adjusted R2 value, indicating a strong
ability of the models to represent the data. Most notably, Bangalore has a significant increase in the
accuracy of the model, with the adjusted R2 value increasing by more than 10 times. This further
proves that the log-linear regression model is a plausible model for rent prices in general.
Table 6. Comparison of adjusted R2 of each city with and without logarithmic transformation
City Adjusted R2 with log Adjusted R2 without log
Kolkata 0.294267 0.53294205
Mumbai 0.749125 0.73906754
Bangalore 0.066323 0.73226164
Delhi 0.491262 0.68253842
Chennai 0.356148 0.66305786
Hyderabad 0.577645 0.67094438

5. Discussion
The results from analyzing the R2 value and p-value are slightly different. The p-values show that
all the factors have a strong correlation with housing rent while R2 indicates that two of the factors,

581
Highlights in Science, Engineering and Technology TPCEE 2022
Volume 38 (2023)

furnishing status, and floor level do not show close relationships with housing rents. This means that
the model or the data we analyzed needs to be improved. We suggest that the difference in the two
analyses is caused by the influence of extreme values. The standard errors shown in Table 2 are very
large, which means that our data is highly discrete and the final results will be highly affected by
extreme values. To improve our model results, it is better to cull out outliers and extreme values.
Moreover, while the models are fairly accurate according to the R2 value, there is no doubt that
there is still room for improvement. The R2 values are still not high enough for the model to be
considered reliable. There might be a few solutions to this. Firstly, removing these outliers can make
our model more accurate, as previously mentioned. Using other advanced modeling methods, such
as gradient boosting and ridge regression, can help make the model more accurate as well.
Our model also fails to account for different areas within a city. This may also be a potential point
of improvement as intuitively location should have a non-trivial impact on house rents.

6. Conclusion
In conclusion, all six factors (number of BHK, size, floor level, number of bathrooms, furnishing
status and city) have a certain effect on housing rents. Among them, in the same location, the number
of bathrooms and size of the house has the largest impact on the rents while floor level and furnishing
status have the smallest influence. Mumbai has the highest housing rent and correlation with different
factors, but Bangalore has the lowest one. In other words, if the buyer's budget is limited, they should
consider a house with a small number of bathrooms or smaller areas in Bangalore. Choosing a
furnished house can help the buyer save some costs to a certain extent as well as they would not need
to spend extra money on furnishing.

References
[1] Dökmeci, Vedia, et al. “External Factors, Housing Values, and Rents: Evidence from Survey Data.”
Journal of Housing Research, vol. 14, no. 1, 2003, pp. 83–99. JSTOR,
[Link] Accessed 6 Oct. 2022.
[2] Ming, Yue, et al. “Prediction and Analysis of Chengdu Housing Rent Based on XGBoost Algorithm.”
Proceedings of the 2020 3rd International Conference on Big Data Technologies, 2020,
[3] Li, Jinze. “Monthly Housing Rent Forecast Based on LightGBM (Light Gradient Boosting) Model.”
International Journal of Intelligent Information and Management Science, vol. 7, no. 6, Dec. 2018, pp.
58–65.
[4] Ma, Tao, et al. “A Study on House Rent Prediction Based on Ensemble Learning.” Finance, Chinese
Academic Journal, Hans Publishers, 25 Oct. 2019.
[5] Kumar, Adarsh. “House Rent Price Prediction.” International Research Journal of Engineering and
Technology, vol. 06, no. 04, Apr. 2019, pp. 3188–3191.
[6] Mateusz Tomal. “Modelling Housing Rents Using Spatial Autoregressive Geographically Weighted
Regression: A Case Study in Cracow, Poland.” International Journal of Geo-Information, May. 2020.
[7] Colema, Andrew, et al. "A Simple Model of Housing Rental and Ownership with Policy Simulations,"
Working Papers 09_08, Motu Economic and Public Policy Research.
[8] Workspace – [Link], [Link]
[9] A., Christina. “8 Factors Affecting Rent Prices.” LinkedIn, 5 Feb. 2021,
[Link]
[10] Bogdan. “International Study: Renting Landscape in 30 Countries around the World.” RentCafe Rental
Blog, 13 May 2021, [Link]
around-world/

582