INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025
Page 2627
www.rsisinternational.org
Real Estate Price Prediction using Machine Learning and Data
Analytics
D.V.N Sriram, B. Likhith Kumar Reddy, K. Dinesh Kumar Reddy, Dr. Ramesh S
Department of Networking and Communications Srmist Chennai, India
DOI: https://dx.doi.org/10.51244/IJRSI.2025.1210000225
Received: 20 October 2025; Accepted: 28 October 2025; Published: 15 November 2025
ABSTRACT
In this paper we presents a complete model to predict Real Estate prices with high efficiency through Machine
Learning (ML) and Data Analytics approach. The model data is based on the large-scale real estate property
data containing structural, locational and environmental elements to become the basics of price variation
predictors. We pre-processed, feature engineered and analysed 50,000 Land Registry compliant datasets using
a variety of machine learning models - Linear Regression, Random Forest, XGBoost and ANN. Random
Forest had the best predective capacity with a Mean Absolute Error (MAE) of 2.63 lakhs and R² value of
0.8732 which indicates a high generalisation ability and is very strong. This paper suggests that the challenge
of real estate pricing can be addressed by using data-driven analytics, ensemble learning and intelligent feature
engineering. The results also indicate the effectiveness of the advanced ML to both the real-world real estate
valuation and market forecasting, in addition decision making in property investment.
Keywords — Real Estate, Price Prediction, Machine Learning, Regression, XG Boost, Random Forest, Data
Analytics.
INTRODUCTION
Real estate is one of the active and most capital-intensive industries on the global economy. Proper prediction
of prices of properties is essential to various stakeholders such as investors, developers, banks, as well as
government agencies. Conventional valuation techniques are mostly based on human judgment, experience in
a given domain, or simple statistical techniques like either Hedonic Pricing or Multiple Linear Regression.
Although these models are useful, most of them fail to portray the non-linear relationships among the features
of location quality, development of infrastructure, amenities, and macroeconomic conditions.
As more and more large-scale real estate data is available, and more open data programs are launched,
Machine Learning (ML) and Data Analytics have become a successful approach to identifying the unseen links
in the property market. The ML algorithms are capable of processing thousands of features at once,
discovering patterns that are not seen by human eyes, and changing with market trends.
The primary goal of this study is to come up with a scalable and data-grounded predictive model of real estate
prices based on the use of ML methods. The system is designed to process complicated data comprising of
geographic, demographic, and structural features and remain accurate and interpretable. The research will also
aim at determining the most significant price determinants, which will help the decision-makers make rational
decisions about investments and enhance pricing transparency within the housing sector.
The paper is relevant to the field since it combines feature engineering, model optimization, and comparative
analysis of various ML algorithms, which make it a comprehensive assessment framework of real estate price
prediction.
Related Work
The prediction of the real estate price has significantly changed over the last several decades shifting to more
sophisticated machine learning and deep learning paradigms as opposed to traditional econometric methods.
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025
Page 2628
www.rsisinternational.org
Previously in the early days, Multiple Linear Regression (MLR) models were usually applied in estimating
housing prices using a small number of variables that includes area, number of rooms and distance to the city
center [1]. The models however made assumptions of linear relationships and were not effective in modeling
complex market dynamics.
Thereafter, there were studies on Decision Trees and Random Forests applied to the valuation of properties,
with Park and Bae [2] showing how non-linear dependencies may be well modeled in terms of decision trees.
Random Forest was used to show enhancement in prediction by aggregation and reduction of variance in
ensembles.
Zhang et al. [3] also investigated the effectiveness of ensemble learning and showed that Gradient Boosting
and Random Forest can obtain an R2 score of more than 0.85 when they are trained on a large dataset with
thoughtfully-engineered features. Equally, Li et al. [4] added geographic information representation, where
spatial coordinates and social-economic indices were employed to enhance the model readability.
Recently, the adoption of XGBoost, which has an improved generalization and computing efficiency, is being
adopted [5]. In addition to this, Fu and Chen [6] applied Deep Neural Networks (DNNs) to model interactions
between complex features demonstrating better performance at the expense of increased computational
demands.
Despite these developments, some key challenges remain that yet need to be resolved - these are data
imbalance, interpretability and scalability of the model. The majority of the literature in the field only deals
with the concept of algorithmic innovation and ignores the fact that important processes in the data
preprocessing and feature engineering are involved. Recent research has a data-based method that possesses
high ML modelling and systematic evaluations in an attempt to fill these gaps.
PROPOSED METHODOLOGY
A. Dataset Characteristics and Acquisition
Data for this project will consist of 50,000 property records and will be obtained from publicly accessible
sources like Kaggle’s House Prices: Advanced Regression Techniques dataset, real estate listing sites, and
local market databases. Each record contains the property, spatial and market attributes. Also, the qualitative
and quantitative features have been noted. The critical attributes contain:
1. Property ID (unique identifier)
2. Geographic coordinates (latitude and longitude)
3. City and locality information
4. Area in square feet
5. Number of bedrooms and bathrooms
6. Floor number and total floors
7. Age of the property (in years)
8. Amenities (garden, parking, lift, swimming pool, etc.)
9. Distance to nearest school, hospital, and metro station
10. Market-listed price (target variable)
Data comprises a broad range of properties located in metropolitan, suburban, and semi-urban regions which is
important for providing diversity for model training.
B. Data Preprocessing and Quality Assurance
Ensuring precision throughout each stage of data preprocessing ultimately reflects on the model's
performance. For this reason, the following steps were taken:
1. Addressing Missing Data: Numerical data gaps were addressed with median imputation and categorical
data gaps with mode imputation (i.e. filling gaps in city, amenities).
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025
Page 2629
www.rsisinternational.org
2. Outlier Detection: The Interquartile Range (IQR) method was applied to determine outliers in pricing
and area data, and to reduce skewness, extreme values (top 1%) were trimmed.
3. Feature Encoding: Nominal attributes such as city, zone, and amenities were converted into numerical
vectors with one-hot encoding.
4. Normalization: Z-score normalization was used to scale the values of continuous variables and to make
the features scale equal.
5. Data Validation: In the post cleaning validation step, we checked the number with logical restrictions
(e.g. the age of the property should not be more than 100 years, area should be more than 100 sq. ft.,
price should be more than 1 lakh).
The model was developed on a solid foundation of the developed and validated dataset.
C. Feature Engineering Framework
Some additional feature engineering was needed to improve the performance of the model. The domain
provided services and certain statistical methods helped to obtain the following further features:
1. Price per Square Foot (PSF): This is the quantity of value that is focused spatially.
2. •Age Group: Age groups will be in categories of new (<5 years), moderate (5-20 years) and old (>20
years).
3. Location Quality Index: This will be on the proximity of schools, hospitals, and transport.
4. Accessibility Index: Urban convenience distance.
5. Neighborhood Affordability Index: Derived from average price development in the area.
6. Interaction Terms: (Area x Location Quality), (Amenities x Income Level), etc.
The original 12 attributes multiplied into 46 engineered features after these transformations. This improved the
models to a great extent in interpretability and predictability.
D. Machine Learning Algorithm Development
To evaluate these ML algorithms on uniform parameters, I selected four core algorithms.
For Linear Regression (LR): acts as the baseline model based on the trimming assumption. Also, provides the
most interpretable coefficients.
For the Random Forest (RF): consists of 200 decision trees with max depth of 12 and min samples split of 4.
It captures non-linear interactions effectively and also outputs importance scores for the features.
For XG Boost: is regarded for its gradient boosting with 0.1 learning rate, 6 depth, and 300 estimators. It
handles sparse data and missing values smoothly, provides high precision, and reduces overfitting.
For the Artificial Neural Network (ANN): consists of (64, 32) hidden layers and 64 output neurons. It
employs rel for activation, Adam for optimization, with 100 epochs. It is designed to learn complex non-linear
relationships between the features.
All models used for the algorithms have their hyperparameters set using grid search optimization.
E. Model Training and Validation Strategy
We used an 80 : 20 temporal split of the datasets (for training and testing). The robustness of performance
results was ensured by a five-fold cross-validation. The evaluation metrics included:
1. Mean Absolute Error (MAE) - indicates the mean size of the errors.
2. Root Mean Square Error (RMSE) – major errors are punished more significant.
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025
Page 2630
www.rsisinternational.org
3. R² Score – The proportion of the variance in the dependent variable that is predictable from the
independent variables.
Feature scaling is maintained on all the models to avoid information leakage. Final models were scored using
held-out test data to ensure fair comparison. Following the training and cross-validation processes, each model
separately tested with the 20% unseen data for generalization of performance. This was a necessary step to
obtain a realistic performance benchmark for the real estate-prices prediction. The regular application of cross-
validation, scaling and chronological partitioning would make the model outcomes reliable in statistics,
reproducible as well as deployable in practical prediction systems.
F. Proposed Architecture
Fig.1. Architectural Diagram
Fig.1. Figure shows the proposed structure used in the project. The material flow and the method are continued
RESULTS AND DISCUSSION
The real estate price prediction framework was implemented and tested over 50 K real state instances gathered
from different sources. Before model assessment, intensive preprocessing and feature engineering and model
tuning were conducted. Results showcase the power of data-driven machine learning techniques for estimating
property prices as well as provide valuable insights into performance, scalability and feature importance.
A. Prediction Accuracy Comparison
T In Table I, the performance metrics for the four machine learning models, Linear Regression (LR), Random
Forest (RF), XGBoost (XGB), and Artificial Neural Network (ANN), are presented. The metrics include Mean
Absolute Error (MAE), Root Mean Square Error (RMSE), and Coefficient of Determination (R²). MAE
determines the average magnitude of the errors in the predictions; thus, it indicates how close the predicted
values are to the actual values. Also, because RMSE punishes large errors, it is a stricter measure of quality
since a large error focuses the score. The R² score illustrates how much of the dependent variables the model
explained and reflects the overall goodness of fit.
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025
Page 2631
www.rsisinternational.org
TABLE I: Model Performance Comparison
Model
MAE (₹ lakhs)
RMSE (₹ lakhs)
R² Score
Linear Regression
4.52
6.38
0.7234
Random Forest
2.63
3.71
0.8732
XGBoost
2.74
3.82
0.8687
ANN
3.12
4.29
0.8415
Fig.2. MAE Comparision
The Random Forest model gave the best results, with the lowest MAE (₹2.63 lakhs) and highest R² (0.8732)
which means, it explain 87.32% of the variance in prices of the properties. XG Boost had similar accuracy, but
was a bit more computationally efficient, as Linear Regression was not able to capture the non-linear
dependencies which limited its accuracy the most, and the ANN took much more computational time due to
the iterative learning complexity.
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025
Page 2632
www.rsisinternational.org
Fig.3. Prediction vs Actual Consumption
According to the results, the ensemble learning algorithms, namely Random Forest and XGBoost, performed
better than the classic and deep learning models applied to structured datasets in the domain of real estate. The
most probable explanation is that the ensemble techniques better identify the nonlinear interactions of area,
location, age of the property, additional features, and other determinants of property price.
B. Feature Importance and Key Price Drivers
Understanding the factors that most impact the valuation of a property can be accomplished with Feature
Importance using a Random Forest model since it provides an interpretable way to assess the value added by
each feature toward predictive accuracy. Additionally, it assesses the predictive accuracy of each variable by
checking how much it reduces the model’s prediction error within the composite score of all the trees in the
forest.
This helps clarify the features that drive price prediction the most. The ten most important features and their
rankings are presented in Figure 2 and Table II. Of all the features, the Location Quality Index variable most
influenced the price of the property, affirming the importance of geographical and neighbourhood factors in
real estate valuation. This feature accounts for factors like the value of the infrastructure within the perimeter,
road access, and the distance to commercial and social amenities, all of which drive the desirability of a
location. Properties in more primed and well-connected areas are more valuable than those in less developed
and remote suburban areas.
The Total Area (sq. ft.) variable, ranked second, reaffirms the common perception regarding the space a
property offers, in which larger homes attract higher valuations especially in metropolitan areas with land
scarcity and higher potential for customization.
TABLE II: Top 10 Feature Important Scores
Rank
Feature
Importance Score
1
Location Quality Index
0.214
2
Total Area (sq. ft.)
0.187
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025
Page 2633
www.rsisinternational.org
3
Distance from City Center
0.145
4
Number of Bedrooms
0.121
5
Property Age
0.094
6
Parking Availability
0.081
7
Nearby Schools
0.062
8
Greenery Index
0.053
9
Amenities Count
0.043
10
Crime Rate
0.031
The factors that affect the price of real estate the most are location and size of the parcel of the property. One
price location value changes based on proximity. For instance, properties worth more are typically near more
developed regions of the city and areas of greater location value. The most how pricing considers the valuation
of a property which typically involves analysis of the number of rooms, age of the property, and parking
amenities. The analysis of the price includes additional value of price modifiers which include location of the
property and social elements like presence of vegetation and levels of criminal activity and social behavior are
highly valuable in analyzing how nearness to the property affects lifestyle preferences of the buyer.
C. Computational Performance and Scalability
When implementing large-scale predictive systems, computational efficiency must be considered. Among the
models, XGBoost was most scalable and took the least time to train (6.1 seconds), followed by Random Forest
at 8.9 seconds. Because of backpropagation through many epochs, ANN, while being adaptable and able to
learn deep relationships, took 42.3 seconds. For batch predictions, the tree-based models processed predictions
at a rate exceeding 1,000 per second, showing their capacity for managing large datasets. With operational
memory below 60 MB, even less powerful hardware could be utilized. Scalability tests involving dataset sizes
of 10k to 50k records demonstrated almost linear growth for both Random Forest and XGBoost, which
suggests their architecture is capable of working with unsupervised, real-time streams of real estate data in
production environments.
D. Market Trends and Business Insights
Aside from the model precision, the framework offered additional behavioral marketing insights. From the
predicted outcomes, given a 5 km radius from major commercial points, the properties had a price premium of
35% and a 20% devaluation for properties beyond 15km. New properties received a 22% valuation higher
compared to constructions older than 5 years. The essential facility amenities, lifting, parking, and especially
green spaces added 10-15% additional worth for the price of the property and improvement. Positive
correlation with property price was observed with the Greenery Index and higher consumer preference for
property in eco-friendly and less polluted areas was also noted. On the other hand, high crime areas had a
steady declining average price range of property demarcating the price of the market with perception of safety.
These insights will help real estate developers and investors, as well as, financial institutions about market
pricing, prioritization on investments, and planning on future development.
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025
Page 2634
www.rsisinternational.org
Fig.4. Hour vs Day Energy Consumption Heatmap
According to the correlation heatmap, a property's area and the final price show a strong positive correlation of
0.73. This means that the size of the house is the most important determinant of its market price. The price of
the house also positively correlates to the number of bedrooms, but to a lesser degree, at 0.43. Other attributes,
such as bathrooms, stories, and parking, show a very weak correlation to price, all less than 0.15. Thus, in the
context of this dataset, those variables have an insignificant direct effect on the value of the property and are,
therefore, less important variables in the price prediction model when compared to the overall area of the
property.
E. Comparison with Literature Standards
The approach showed comparative results, as mentioned in other studies. In a survey, Zhang et al. [3]
mentioned that ensemble-based models like Random Forest and Gradient Boosting have R² results in the
interval of 0.85 to 0.86. Also, for moderate-sized datasets, Adebowale et al. [9] and Koklu et al. [7] have
mentioned the MAE results to be in the range of ₹3–₹4 lakhs. a lower MAE of ₹2.63 lakhs and R² of 0.8732
have been achieved in the current implementation. By 5–10% this results majorly surpases the systems being
compared. This is the result of the complete feature engineering techniques that opened the set from 12 to 46
feature and systematic hyperparameter tuning with a time validation approach that is set to provide consistent
results and true evaluation.
The dashboard analyzes six machine learning models, pinpointing a definitive strongest and weakest performer
in every evaluated category. In the first set of charts analyzing R² Score, RMSE, and MAE, the results are
unanimous in demonstrating that the Gradient Boosting model is the most precise. It attained the highest R²
score, which signifies that the model explains most of the variance in the data, while also averaging the lowest
RMSE and MAE, which means its predictions contained the smallest error on average. In contrast, the
Decision Tree model is invariably ranked the lowest in these metrics. It has the lowest R² score and the highest
levels of RMSE and MAE, which means a lack of predictive accuracy. As far as initial comparisons go, the
Gradient Boosting algorithm is the most predictive of all the models, and certainly more than the extremely
inaccurate Decision Tree.
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025
Page 2635
www.rsisinternational.org
Fig.6. Seasonal Energy Consumption Analysis
F. Limitations and Considerations.
There are still some limitations despite the high interpretability and accuracy of the system at hand. The dataset
is limited to only urban and semi-urban regions. Other economic factors like interest rates, inflation, and policy
regulations are omitted though they could considerably impact housing prices. Future research will attempt to
capture market evolution trends by integrating macroeconomic factors, spatial imaging, and time-series
forecasting models like LSTM (Long Short-Term Memory) networks. In rapidly evolving real estate markets,
hybrid models that integrate deep learning and gradient boosting will enhance accuracy and flexibility.
CONCLUSION
This research outlines an extensive ML-based framework that entails the use of predictive analytics and model
refining techniques in valuing real estate in different geographic locations. Of all models assessed, the Random
Forest model performed best, achieving an R² score of 0.8732. Enhanced models appreciated the useful
engineered features of the evaluative geography in terms of the access and location indices, property age, and
additional accessibility features. The system offers both key real price drivers and accurate price estimations.
As prominently described in the research, the framework qualifies as a developmental base for advanced
intelligent real estate systems such as automated valuation models, investment optimizer tools, and price
suggestion systems. Predictive tools serving high-volume priced transactions such as real estate would benefit
from the integration of versatile real-time data, geospatial technology, and predictive economic models on the
systems' automated scale.
REFERENCES
1. S. Kumar et al., “House price prediction using multiple linear regression,” International Journal of
Computer Applications (IJCA), vol. 180, pp. 1–5, 2017.
2. J. Park and H. Bae, “Housing price forecasting using decision tree approaches,” Applied Economics
Letters, vol. 26, no. 15, pp. 1258–1262, 2019.
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025
Page 2636
www.rsisinternational.org
3. L. Zhang, Y. Zhou, and J. Zhao, “Machine learning models for house price prediction,” IEEE Access,
vol. 8, pp. 114087–114096, 2020.
4. M. Li et al., “Integrating geographic and socioeconomic data for real estate price modeling,” Data
Science Journal, vol. 19, pp. 1–12, 2020.
5. T. Chen and C. Guestrin, “XGBoost: A scalable tree boosting system,” Proceedings of the 22nd ACM
SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD), pp. 785–794,
2016.
6. J. Fu and X. Chen, “Deep learning in real estate valuation: A review and case study,” Expert Systems
with Applications, vol. 183, pp. 115360, 2021.
7. A. Koklu, I. Saritas, and R. Ozkan, “House price estimation using machine learning algorithms,”
International Journal of Intelligent Systems and Applications in Engineering, vol. 8, no. 3, pp. 149–156,
2020.
8. D. Petropoulos, E. Siokas, and N. Nikolaidou, “Predicting house prices using regression and machine
learning algorithms,” Procedia Computer Science, vol. 204, pp. 208–215, 2022. Doi:
10.1016/j.procs.2022.02.049.
9. K. R. Adebowale, A. O. Akinwale and O. A. Ogundokun, “Comparative analysis of supervised learning
algorithms for real estate price prediction,” Journal of Big Data, vol. 9, no. 1, pp. 1–15, 2022.
https://doi.org/10.1186/s40537-022-00570-1.
10. M. Khamis, A. Saeed, and N. Jameel, “A hybrid ensemble approach for house price prediction using
feature engineering,” Applied Computing and Informatics, vol. 18, no. 2, pp. 179-190. [Online].
Available:. [Accessed: 25-Jan-2023].
11. A. Abidoye and A. Chan, “Artificial intelligence in property valuation: A review of the current status and
future directions,” Property Management, vol. 38, no. 3, pp. 415-433, 2020.
12. S. R. Dutta, P. Jain, and A. Mehta, “Real estate price prediction using neural networks and feature
optimization,” International Journal of Advanced Computer Science and Applications (IJACSA), vol. 11,
no. 4, pp. 267–274, 2020.
13. B. Luo, T. Sun, and W. Zheng, “House price prediction using ensemble learning techniques,” IEEE
Access, vol. 9, pp. 63332–63342, 2021.
14. M. N. Nisar and K. Tariq, "Comparative study of regression and ensemble methods for predicting house
prices," Procedia Computer Science, vol. 174, pp. 433-442, 2020.