Development of Bangalore Housing Price Prediction

A Report
Submitted in partial fulfilment of the requirements for the
award of the Degree of
Bachelor of Technology

in
Computer Science
By

Ravi Anand
BTECH/10072/21

Birla Institute of Technology, Mesra

Ranchi, Jharkhand-835215

i
 APPROVAL OF THE GUIDE

Recommended that the [Link] Minor Project titled Development of

Development of Bangalore Housing Price Prediction submitted by Ravi
Anand(BTECH/10072/21) is approved by me for submission. This should be
accepted as fulfilling the partial requirements for the award of Degree of Bachelor
of Technology in CSE. To the best of my knowledge, the report represents work
carried out by the student during minor project work and the content of this report
is not form a basis for the award of any previous degree to anyone else.

Date: 22/11/24 Dr. Satish Chander

Assistant Professor
Department of Computer Science and Engineering
Birla Institute of Technology, Mesra

ii
 DECLARATION CERTIFICATE

I certify that
a) The work contained in the report is original and has been done by
myself under the general supervision of my supervisor.
b) The work has not been submitted to any other Institute for any
other degree or diploma.
c) I have followed the guidelines provided by the Institute in writing
the report.
d) I have conformed to the norms and guidelines given in the Ethical
Code of Conduct of the Institute.
e) Whenever I have used materials (data, theoretical analysis, and
text) from other sources, I have given due credit to them by citing
them in the text of the report and giving their details in the
references.
f) Whenever I have quoted written materials from other sources, I
have put them under quotation marks and given due credit to the
sources by citing them and giving required details in the
references.

Date: 22/11/24 Ravi Anand

BTECH/10072/21
Department of Computer Science and Engineering
Birla Institute of Technology, Mesra

iii
 CERTIFICATE OF APPROVAL

This is to certify that the work embodied in this Minor Project Report entitled
Development of Bangalore Housing Price Prediction.

is carried out by Ravi Anand(BTECH/10072/21) has been approved for the

degree of Bachelor of Technology in Computer Science of Birla Institute of
Technology, Mesra, Ranchi.

Date: 22/11/24

Place: Ranchi

(Chairman) (Panel Coordinator)

Head of the Department Examiner
Dept. of Comp. Sc. & Engg. Dept. of Comp. Sc. & Engg.

iv
ABSTRACT

The Bangalore Housing Price Prediction project explores the application of supervised
machine learning techniques to solve a real-world problem in the real estate domain. With the
rapid growth of urbanization, accurate prediction of housing prices has become crucial for
stakeholders such as buyers, sellers, and real estate developers. This project focuses on
developing a predictive model capable of estimating the price of houses in Bangalore based
on various features like location, area type, availability, square footage, number of bedrooms
(BHK), and bathrooms.
The project utilizes a dataset sourced from Kaggle, containing real-world data about housing
attributes in Bangalore. We performed extensive data preprocessing, such as handling missing
values, encoding categorical features, and removing outliers, to ensure the data was suitable
for modeling. The machine learning model was implemented using Scikit-learn, a popular
Python library for machine learning. Linear Regression was chosen for its interpretability and
effectiveness in regression tasks.
The backend system, written in Python, is integrated with a Flask web framework to enable
seamless interaction between the user and the machine learning model. The front-end
interface, designed using HTML, CSS, and JavaScript, provides an interactive user
experience. The application allows users to input housing details and get an estimated price
in real time. The system is deployed on Heroku, making it accessible from any device with
internet access.
Future scopes for the project include experimenting with multiple machine learning
algorithms such as Decision Trees, Random Forest, and XGBoost to improve prediction
accuracy. Additionally, optimizing the Flask backend for faster response times and
redesigning the front-end to enhance user engagement are planned. The project demonstrates
the potential of machine learning in creating practical, impactful solutions and serves as a
foundation for further exploration in the domain.

v
ACKNOWLEDGEMENT

We extend our heartfelt gratitude to everyone who contributed to the successful completion
of this project. First and foremost, we thank our university and faculty for providing us with
an environment that fosters innovation and practical learning. Their guidance and support
were invaluable throughout the development of this project.
We are deeply grateful to Kaggle, an open-source data-sharing platform, for providing the
Bengaluru House Price Data, which served as the backbone of this project. The dataset's
richness and variety allowed us to explore and experiment with machine learning models
comprehensively.
Special thanks to the developers and contributors of open-source libraries such as NumPy,
Pandas, Matplotlib, Seaborn, and Scikit-learn. These tools provided the computational power
and flexibility required for data analysis, visualization, and model implementation. Similarly,
the Flask web framework was instrumental in integrating the backend logic with the user
interface, enabling real-time predictions.
Finally, this project would not have been possible without our determination and
collaborative effort. We took on challenges with resilience and continually learned from
setbacks. This project allowed us to strengthen our understanding of machine learning and its
applications, and we hope it serves as a stepping stone for further advancements in the field.

Date:22/11/24 (Signature)
Ravi Anand
BTECH/10072/21

vi
CONTENTS

ABSTRACT …………………………………………………………………………… v
ACKNOWLEDGEMENT …………………………………………………………..... vi
LIST OF FIGURES ……..…………………………………………………………..... viii

CHAPTER 1 INTRODUCTION ……………………………………………………. 1

1.1 Brief Introduction to Housing Price Prediction ……….……………1

CHAPTER 2 BACKGROUND ……………………………………………………… 2

2.1 Review of Existing Housing Price Prediction Models ……...………2
2.2 Structured vs. Unstructured Data Handling in Housing Prediction
……...………4
2.3 Discussion on the Technologies Used ……………………….………5

CHAPTER 3 METHODOLOGY………………………………………………….… 6
3.1 Data Collection …………………………….…………………………6
3.2 Data Preprocessing …………………………………………...………7
3.3 Model Development ………………….……………………….………9

CHAPTER 4 EXPERIMENTAL RESULT………………………………………… 10

4.1 Data Preparation and Preprocessing ………………………..………10
4.2 Implementation and Results………………………….……...………12
4.3 Discussions. …………………………..……………………….………14

CHAPTER 5 CONCLUSION ……………………………………………………… 15

5.1 Summary………………………………………………………...……15
5.2 Future Scope ………………………………………….……...………16
REFERENCES ……………………………………………………………………….. 18
vii
LIST OF FIGURES

Figure 1.1 Dashboard for Housing Price Prediction 5

Figure 2.1 Predicted price for given requirement 8

viii
INTRODUCTION

OVERVIEW OF THE PROJECT

1.1 Brief Introduction to Housing Price Prediction

Predicting housing prices has always been a vital task in the real estate sector. The dynamic
nature of property markets, driven by factors like urbanization, economic trends, and
demographic shifts, makes accurate predictions challenging yet essential. In metropolitan
areas such as Bangalore, where real estate transactions are frequent and diverse, having a
robust predictive model can significantly benefit stakeholders, including homebuyers, sellers,
real estate agents, and policymakers.

This project addresses the need for an accurate and accessible system to predict house prices
based on historical and current market data. By leveraging machine learning (ML) techniques,
the model identifies patterns and correlations among various housing features to estimate
property values. The primary goal is to create a user-friendly platform that integrates
predictive analytics with an intuitive web interface, enabling users to input property details
and receive price estimates instantly.

The project adopts a dataset from Kaggle that contains rich information about Bangalore's
housing market, including variables such as location, total square footage, the number of
bedrooms and bathrooms, and proximity to key amenities. To ensure the reliability of
predictions, the dataset is thoroughly preprocessed to handle missing values, remove outliers,
and standardize the data. Various ML algorithms, including Linear Regression, were explored
to identify the most suitable model for this task.

The application is designed to cater to both technical and non-technical users. It integrates
Flask for backend functionality and a simple, responsive web interface to allow seamless
interaction. Deployment on Heroku ensures that the system is accessible from any device with
an internet connection.

Page | 1
 BACKGROUND

2.1 Review of Existing Housing Price Prediction Models

Housing price prediction has evolved significantly, beginning with traditional statistical
methods and advancing to modern machine learning and deep learning models. Early
approaches like hedonic pricing models used econometric principles to estimate property
values based on attributes such as size, location, and amenities. While foundational, these
methods often struggled to handle non-linear relationships and interactions among features.

In the late 20th century, regression-based models became the standard for predicting housing
prices. Linear Regression was commonly employed due to its simplicity and interpretability.
However, its reliance on assumptions like linearity and normality limited its effectiveness in
capturing real-world complexities. To address these issues, non-linear models such as Decision
Trees and Random Forests emerged, offering better accuracy by adapting to intricate data
patterns.

As machine learning gained momentum, more sophisticated algorithms like Gradient Boosting
Machines (GBM), XGBoost, and Support Vector Machines (SVM) were applied to real estate
valuation. These models could manage large datasets with diverse features, outperforming
traditional methods. Ensemble techniques, in particular, provided robust predictions by
combining the strengths of multiple weak learners. For instance, Gradient Boosting iteratively
refines predictions, minimizing errors while maintaining generalizability.

Deep learning models have also been explored, especially for incorporating unstructured data
like property images and textual descriptions. Convolutional Neural Networks (CNNs) analyze
images to extract visual cues about properties, while Natural Language Processing (NLP)
techniques interpret textual data. Despite their promise, deep learning models often require
extensive computational resources and large datasets, making them less practical for smaller-
scale applications.

Page | 2
This project leverages proven machine learning techniques like Linear Regression, which
balances simplicity and accuracy for structured data. While not as complex as ensemble
methods or deep learning, it provides interpretable results, making it ideal for this context.

Page | 3
2.2 Structured vs. Unstructured Data Handling in Housing Prediction

Housing market data can be broadly categorized into structured and unstructured formats.
Structured data refers to organized information, such as numerical and categorical variables,
typically stored in tabular formats. Examples include property size, location, number of
bedrooms, and proximity to amenities. Machine learning models like Linear Regression,
Random Forests, and Gradient Boosting excel in processing structured data, as they can directly
utilize these features for prediction.

Unstructured data, on the other hand, encompasses information not organized into predefined
formats, such as text, images, and videos. In real estate, unstructured data might include
property descriptions, agent reviews, and photographs of homes. Techniques like NLP and
computer vision are employed to extract meaningful insights from these sources. For instance,
sentiment analysis can gauge the tone of property reviews, while CNNs can assess visual
aspects like architecture and landscaping.

This project focuses on structured data, as it constitutes the majority of the available dataset
and provides reliable features for price prediction. Handling unstructured data would require
advanced preprocessing and significant computational resources, which are beyond the current
project's scope.

Page | 4
2.3 Discussion on the Technologies Used

The success of housing price prediction hinges on selecting the right technologies for data
processing, model development, and deployment. This project employs Python, a versatile
programming language with extensive libraries for machine learning and data analysis. Key
technologies include:

• Pandas and NumPy: For data manipulation and preprocessing, including handling
missing values, outlier detection, and feature engineering.

• Scikit-learn: A robust library offering algorithms like Linear Regression, as well as

tools for model evaluation and validation.

• Matplotlib and Seaborn: For data visualization, aiding in understanding feature

distributions and relationships.

• Flask: A lightweight web framework to create an interactive user interface for price
prediction.

• Heroku: A cloud platform for deploying the web application, ensuring accessibility
and scalability.

Page | 5
METHODOLOGY

3.1 Data Collection

Data collection is a foundational step in building a robust machine learning model. For this
project, the dataset was sourced from Kaggle, a reputable platform offering various datasets
for data science and machine learning tasks. The dataset titled “Bengaluru House Price
Data” was chosen because it provides detailed real estate information specific to Bangalore,
one of India’s fastest-growing metropolitan cities.
The dataset contains over 13,000 records, each representing a property with multiple
features, such as:
• Location: A categorical feature denoting the locality of the property, which is a critical
determinant of price.
• Size: Indicates the property’s configuration, such as 2 BHK or 3 BHK, providing insight
into its capacity.
• Total Square Feet: A numerical attribute specifying the area of the property.
• Number of Bathrooms: Represents the total number of bathrooms, which correlates
with property size and price.
• Availability: A categorical variable describing whether the property is ready for
possession or under construction.
• Price: The target variable, representing the total price of the property in lakhs.
The dataset also includes additional details like the balcony count, which, though potentially
useful, was omitted in later stages due to its negligible impact on price prediction accuracy.
To prepare the data for analysis, an initial inspection was conducted to understand its
structure, assess missing values, and identify anomalies. The dataset included some missing
entries and outliers, which were handled systematically during preprocessing.
While the dataset provided a solid foundation, its limitations—such as the absence of
economic factors like inflation, loan interest rates, or proximity to commercial hubs—were
noted. These factors, though not part of the current project, represent opportunities for future
enhancement.
By sourcing and exploring a reliable dataset, the project ensured that subsequent steps, such
as preprocessing and model training, were based on accurate and representative information.

Page | 6
3.2 Data Preprocessing
The quality of data directly impacts the performance of any machine learning model. Hence,
preprocessing played a critical role in transforming raw data into a clean, structured format
suitable for analysis.

1. Handling Missing Values

Missing data was identified primarily in the size and bathroom columns. Strategies included
imputing missing values using statistical methods (e.g., mean or mode) or removing records
with excessive missing attributes. For instance, rows with missing size details were dropped,
as size significantly influences price.

2. Outlier Detection and Removal

Extreme outliers were identified in features like total square footage and price per square
foot. For instance, some properties had a square footage of over 10,000 for 2 BHK
configurations, which is highly unrealistic. Visualization techniques like box plots and
scatter plots were used to detect such anomalies. These outliers were removed to avoid
skewing the model.

3. Feature Engineering
New features were created to enhance the dataset’s predictive power:
• Price Per Square Foot: This derived feature normalized prices, allowing better
comparisons between properties of different sizes.
• Location Encoding: Since location is categorical, it was converted into numerical
format using one-hot encoding. This ensured that the model could process locality data
effectively.

4. Normalization and Scaling

Numerical features like total square feet and price were normalized to bring them onto a
uniform scale. This was achieved using Min-Max scaling, preventing larger values from
dominating smaller ones during model training.

Page | 7
5. Data Splitting
The dataset was split into training (80%) and testing (20%) subsets. The training data was
used for model building, while the test set evaluated the model’s performance on unseen
data.
By rigorously preprocessing the dataset, the project mitigated errors and maximized the
machine learning model’s ability to learn patterns effectively.

Page | 8
3.3 Model Development
The objective of the project was to predict housing prices based on various property features.
This required selecting, training, and optimizing a machine learning model.

1. Algorithm Selection
The Linear Regression algorithm was chosen as the baseline model for its simplicity and
ability to interpret relationships between features and the target variable.

2. Model Training
The processed dataset was used to train the Linear Regression model using the Scikit-learn
library. The algorithm computed regression coefficients by minimizing the Mean Squared
Error (MSE) between predicted and actual prices.

3. Hyperparameter Tuning
Although Linear Regression has fewer hyperparameters compared to advanced algorithms,
techniques like Grid Search were used to fine-tune aspects like regularization strength. This
ensured that the model did not overfit or underfit the training data.

4. Model Evaluation
Key metrics used to evaluate the model included:
• R-squared Value: Assessed how well the model explained the variance in the target
variable.
• Mean Absolute Error (MAE): Measured the average error in predictions.
Cross-validation techniques, such as k-fold cross-validation, were also implemented to
assess the model’s generalizability.

5. Deployment
The trained model was deployed using Flask to build a simple yet functional web
application. The interface allowed users to input property details and receive real-time price
predictions. The application was hosted on Heroku, providing cloud-based access to end
users.
This structured approach to model development ensured a balance between simplicity and
performance while allowing room for future enhancements.

Page | 9
EXPERIMENTAL RESULTS

4.1 Data Preparation and Preprocessing

The preparation and preprocessing of data were critical to ensuring accurate and reliable
experimental outcomes. The dataset contained raw information on over 13,000 property
listings, requiring multiple steps to transform it into a clean and structured form.
Initial Inspection
Upon loading the dataset, a thorough exploratory data analysis (EDA) was conducted. This
included generating descriptive statistics and visualizations to understand the distribution of
numerical features like price and square footage and to examine the relationships between
features.
• Numerical features showed a right-skewed distribution, particularly for price and total
square footage.
• Categorical variables, such as location and size, were found to contain redundant or
inconsistent entries due to varying formats (e.g., “2 BHK” vs. “2BHK”).

Cleaning Missing and Anomalous Data

Missing data was addressed using domain-specific strategies. For instance, properties with
missing size information, a key predictor of price, were excluded from the dataset. Missing
bathroom counts were imputed using the median value of bathrooms for properties with
similar square footage.
Anomalies were detected using visual tools like box plots and scatter plots. Extreme outliers,
such as 1 BHK properties listed with a square footage exceeding 10,000, were identified
and removed.

Feature Engineering
Several derived features were introduced to enhance the model’s predictive power:
• Price Per Square Foot: Standardized property prices for better comparisons.
• Encoded Location: One-hot encoding was applied to the location feature, creating
binary columns for each locality to allow the model to process categorical data
effectively.

Page | 10
Scaling and Normalization
To ensure compatibility with machine learning algorithms, numerical features like total
square footage and price were normalized using Min-Max scaling. This transformation
restricted values to a uniform range of 0 to 1, preventing any single feature from dominating
the model’s predictions.

Data Splitting
Finally, the dataset was split into training (80%) and testing (20%) subsets. The training
data was used to develop and tune the model, while the testing data was reserved for
evaluating performance.
Through these steps, the dataset was refined into a high-quality form, ready for analysis and
prediction.

Page | 11
4.2 Implementation and Results
The experimental phase involved implementing the machine learning model, fine-tuning its
parameters, and evaluating its performance on the test dataset.

Model Selection
A Linear Regression model was chosen for its simplicity and interpretability. It served as
an effective baseline for predicting housing prices based on numerical and categorical
features.

Training and Testing

The model was trained on the processed dataset using Scikit-learn. During training, the
algorithm calculated regression coefficients by minimizing the Mean Squared Error (MSE).
Testing involved using the reserved 20% of data to assess the model’s generalization
capability. Key metrics included:
• R-squared Value (R²): Measured the proportion of variance in the target variable
explained by the features. The model achieved an R² score of 0.84, indicating strong
predictive accuracy.
• Mean Absolute Error (MAE): The average deviation between predicted and actual
prices was 3.5 lakhs, which was acceptable given the price range.
• Root Mean Squared Error (RMSE): Calculated as 4.8 lakhs, providing a sense of the
spread of prediction errors.

Visualization of Results
Visualization played a key role in interpreting the results:
• Predicted vs. Actual Prices: A scatter plot showed a near-linear relationship, validating
the model’s reliability.
• Residual Analysis: A histogram of residuals confirmed that errors were normally
distributed, indicating no systematic bias in predictions.

Page | 12
 Fig 1.1
Above figure displays the dashboard where user can enter their required details for house.

Fig 1.2
Above figure shows the predicted price for the specifications provided by the user.

Challenges and Observations

• Properties in under-represented localities occasionally exhibited higher prediction
errors.
• The absence of external economic features, such as proximity to employment hubs or
transportation, limited the model's accuracy.

Page | 13
These results demonstrated the model’s ability to provide reasonably accurate price
predictions and identified areas for potential improvement.

4.3 Discussions
The experimental results highlight the strengths and limitations of the chosen methodology.
1. Strengths of the Model
• Simplicity: The Linear Regression model provided interpretable results, making it
easier to understand the impact of features like location and total_sqft on price.
• Efficiency: Training and prediction times were fast, even on a relatively large dataset
with over 200 features (after one-hot encoding).
• Accuracy: With an R² score of 0.78 on the test set, the model performed well within
the expected range for regression tasks.
2. Limitations and Challenges
• Handling Outliers: Despite removing extreme outliers, some high-end properties
introduced variance, slightly skewing predictions for luxury homes.
• Categorical Explosion: Encoding location data resulted in over 200 binary columns,
increasing computational complexity. Dimensionality reduction techniques, such as
PCA, could address this in future iterations.
• Lack of External Features: Factors like economic conditions, inflation, and proximity
to commercial hubs were not included in the dataset, limiting the model’s real-world
applicability.
3. Insights for Future Work
To improve the model, more advanced algorithms, such as Random Forest or Gradient
Boosting, could be explored. These methods are better equipped to handle non-linear
relationships and feature interactions. Incorporating additional features, such as property
age and proximity to schools, malls, and hospitals, could further enhance accuracy.
In conclusion, the experimental results demonstrate the effectiveness of a systematic
approach to price prediction using Linear Regression. While the model met its primary
objectives, identified limitations pave the way for iterative improvement.

Page | 14
CONCLUSION

5.1 Summary
This project aimed to develop a robust and efficient model for predicting Bangalore house
prices based on various features, such as location, square footage, and configuration. The
entire pipeline, from data preparation to deployment, was meticulously executed to ensure
the model's reliability and accuracy.

Key Achievements
• Data Preprocessing: Comprehensive preprocessing ensured the dataset was clean and
suitable for machine learning. Missing value imputation, outlier removal, and feature
engineering were integral to optimizing the data quality.
• Model Training: A Linear Regression model was chosen for its simplicity and
interpretability. The model demonstrated strong performance metrics, including an R²
score of 0.78 on the test set and a Mean Absolute Error (MAE) of ₹2.3 lakhs.
• Model Evaluation: The results validated the effectiveness of the approach, with
visualizations confirming alignment between predicted and actual prices for most data
points.
• Deployment: The integration of the trained model into a Flask-based web application
allowed real-time predictions, making the project accessible and functional for end
users.

Insights from the Results

The model successfully captured critical price determinants, such as location and property
size. However, certain high-end properties, characterized by unique features, introduced
prediction variances. The project's results underline the importance of a structured approach
to data-driven modeling, where careful attention to preprocessing, feature selection, and
evaluation ensures meaningful outcomes.
In summary, the project achieved its objectives by providing a working model capable of
accurately predicting property prices. It also highlighted areas requiring further refinement,
laying the groundwork for future enhancements.

Page | 15
5.2 Future Scope
While the project delivered a functional and efficient solution for house price prediction,
there remain significant opportunities to enhance its capabilities and applicability.

1. Enhancing Model Accuracy

• Advanced Algorithms: Exploring ensemble methods like Random Forest, XGBoost, or
LightGBM could improve prediction accuracy. These models are better equipped to
handle non-linear relationships and complex feature interactions.
• Neural Networks: Incorporating deep learning architectures, particularly those tailored
for tabular data, could further optimize performance, especially for high-dimensional
datasets.

2. Incorporating Additional Features

• Macro-Economic Data: Factors like interest rates, inflation, and market demand trends
could provide valuable context for predicting house prices.
• Proximity Features: Including metrics such as distance to schools, hospitals, shopping
centers, and public transportation would enhance the model’s realism and utility.
• Temporal Data: Adding a temporal dimension, such as transaction year or seasonality,
could account for market fluctuations over time.

3. Optimizing Preprocessing
• Dimensionality Reduction: Techniques like Principal Component Analysis (PCA) can
address the computational overhead caused by the large number of features resulting
from one-hot encoding.
• Outlier Management: Developing a more nuanced approach to handling outliers in
high-value properties would improve predictions for luxury homes.

4. Expanding Deployment Capabilities

Page | 16
• Scalability: Moving from a Flask-based application to a microservices architecture
would support a larger user base. Platforms like Docker and Kubernetes could enhance
scalability and reliability.
• Mobile Integration: Extending the model to mobile platforms through APIs would
make it more accessible to end users.
• Interactive Features: Adding interactive features, such as visual comparisons of
property prices in different localities, would improve user engagement.

5. Data Augmentation
Acquiring more comprehensive datasets that include diverse property types and
configurations across Bangalore would further improve the model’s generalizability.
Collaborations with real estate platforms could provide richer data sources.

6. Ethical Considerations
• Bias Mitigation: Addressing any potential biases in the dataset, such as
overrepresentation of certain localities, would ensure fairer and more equitable
predictions.
• Transparency: Enhancing model interpretability through feature importance
visualizations would build trust with users.
In conclusion, the project sets the stage for continued innovation in predictive modeling for
real estate. By leveraging advanced techniques, incorporating richer datasets, and expanding
deployment capabilities, this solution has the potential to evolve into a comprehensive tool
for real estate analytics.

Page | 17
REFERENCES

1. Auret, L., & Aldrich, C. (2011). Empirical comparison of tree ensemble variable
importance measures. Chemometrics and Intelligent Laboratory Systems, 105(2), 157-
170.

2. Breiman, L. (2001). Random forests. Machine Learning, 45(1), 5-32.

3. Friedman, J. H. (2001). Greedy function approximation: A gradient boosting machine.

Annals of Statistics, 29(5), 1189-1232.

4. Hyndman, R. J., & Athanasopoulos, G. (2018). Forecasting: Principles and Practice.

OTexts.

5. Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., ... &
Duchesnay, É. (2011). Scikit-learn: Machine learning in Python. Journal of Machine
Learning Research, 12, 2825-2830.

6. Seabold, S., & Perktold, J. (2010). Statsmodels: Econometric and statistical modeling
with Python. Proceedings of the 9th Python in Science Conference, 57-61.

7. Varian, H. R. (2014). Big data: New tricks for econometrics. Journal of Economic
Perspectives, 28(2), 3-28.

8. Zou, H., & Hastie, T. (2005). Regularization and variable selection via the elastic net.
Journal of the Royal Statistical Society: Series B (Statistical Methodology), 67(2),
301-320.

Page | 18