St.

Thomas’ College of Engineering and Technology

Synopsis on

Cyclone Casualty and Property Loss Prediction

Department of Information Technology

By

Satarupa Das 12200220034 201220100210029 OF 2020-21

Pratiksha Naskar 12200220055 201220100210008 OF 2020-21

Sayan Das 12200220010 201220100210053 OF 2020-21

Srijit Sarkhel 12200220062 201220100210001 OF 2020-21

Under the guidance of

Mr. Amit Dutta

St. Thomas’ College of Engineering and Technology

Affiliated to

Maulana Abul Kalam Azad University of Technology, West Bengal

November, 2023
St. Thomas’ College of Engineering and Technology

Declaration Page

We are submitting the synopsis on Cyclone Casualty and Property Loss Prediction as a part of our final year
seventh semester project under the guidance of Mr. Amit Dutta

Satarupa Das Pratiksha Naskar Sayan Das Srijit Sarkhel

Guide’s Signature

Mr. Amit Dutta

Department of Information Technology

St. Thomas’ College of Engineering and Technology

Department of Information Technology

Vision
To promote the advancement of learning in Information Technology through research-oriented dissemination of
knowledge which will lead to innovative applications of information in industry and society.

Mission
To incubate students, grow into industry ready professionals, proficient research scholars and enterprising
entrepreneurs.
To create a learner- centric environment that motivates the students in adopting emerging technologies of the rapidly
changing information society.
To promote social, environmental and technological responsiveness among the members of the faculty and student.

PEO:
PEO1: Exhibit the skills and knowledge required to design, develop and implement IT solutions for real life problems.

PEO2: Excel in professional career, higher education and research.

PEO3: Demonstrate professionalism, entrepreneurship, ethical behavior, communication skills and collaborative team
work to adapt the emerging trends by engaging in lifelong learning.

Project Mapping with Program Outcomes:

PO1 PO2 PO3 PO4 PO5 PO6 PO7 PO8 PO9 PO10 PO11 PO12
3 2 1 - - - - 3 2 - 1 -

Enter correlation levels 1, 2 or 3 as defined below:

1: Slight (Low) 2: Moderate (Medium) 3: Substantial (High)

Justification for the Project Mapping with POs:

PO1(Engineering Knowledge): Utilizes engineering knowledge for data processing and modeling.

PO2(Problem Analysis): Involves analysis of problem and available datasets.

PO3(Design/Development of Solutions): Provides a solution with social benefit.

PO8(Ethics): Developed according to professional ethics.

PO9(Individual and Team Work): Team-based project with individual contributions.

PO11(Project Management and Finance): Project management done to complete in time.

Department of Information Technology

St. Thomas’ College of Engineering and Technology

Project Mapping with Program Specific Outcomes (PSOs):

PSO 1 PSO 2

2 1

Enter correlation levels 1, 2 or 3 as defined below:

1: Slight (Low) 2: Moderate (Medium)

Justification for the Project Mapping with PSOs:

PSO1: Programming required for data preprocessing and model implementation.

PSO2: Usage of web technologies for visualizations.

Department of Information Technology

St. Thomas’ College of Engineering and Technology

INDEX

Table of Contents Page No.

Introduction 1
Problem Statement 1
Problem Definition
Objective 2
Literature Survey / Background Study 2
Brief Discussion on Problem 3

Concepts and problem analysis 4

Conclusion 13

Annexure : Reference / Bibliography 14

Department of Information Technology

St. Thomas’ College of Engineering and Technology

1.0. Introduction:
In the face of escalating climate challenges, predicting the impact of cyclones on both human lives and property is of
paramount importance. The field of Cyclone Casualty and Property Loss Prediction employs advanced modeling
techniques and data analytics to anticipate the devastating consequences of these natural disasters. By combining
meteorological data, geographical information, and risk assessment algorithms, researchers strive to enhance our
ability to forecast casualties and property damage caused by cyclones. This interdisciplinary approach not only aids
in early warning systems but also plays a crucial role in developing strategies to mitigate the human and economic
toll of cyclones on vulnerable communities. As the frequency and intensity of cyclones continue to rise, accurate
prediction models become essential for informed decision-making and disaster preparedness.

1.1. Problem Statement:

Design and Develop a Cyclone Casualty and Property Loss Prediction System.

1.2. Problem Definition:

Cyclone Casualty and Property Loss Prediction System consists of an interface where the user can view the
predicted casualty and property loss of an upcoming cyclone. The user gets an approx. idea about the loss that is
going to occur in the specified area entered by the user.

Key Terms:

• Machine Learning - Using algorithms that can learn from data to make predictions

• Feature Engineering - Process of extracting predictive variables from raw data

• Normalization - Rescaling data to a common scale, like 0-1 range

• Training and Testing Sets - Splitting data into subsets to train and evaluate models

• Overfitting - When model fits training data too closely but generalizes poorly

• Underfitting - When model fails to capture key patterns in the data

• Accuracy Metrics - Performance measures like MAE, MSE, R-squared to evaluate models

• Cross-validation - Evaluating model performance across different subsets of data

• Ensemble Models - Combining multiple models to improve overall predictive capability

Department of Information Technology Page | 1

St. Thomas’ College of Engineering and Technology

1.3. Objective:
The key objectives of this project are:

▪ To develop predictive models to forecast cyclone casualties and property damage.

▪ To identify key parameters that influence cyclone impact.
▪ To deliver real-time cyclone impact predictions to facilitate timely response.
▪ To evaluate model performance using appropriate metrics.
▪ To build an interpretable system that can explain the predictions.

1.4 Background Study:

• Evaluation of Tropical Cyclone Disaster Loss Using Machine Learning Algorithms with an eXplainable
Artificial Intelligence Approach[1] by Shuxian Liu, Yang Liu, Zhigang Chu, Kun Yang, Guanlan Wang,
Lisheng Zhang and Yuanda Zhang remarked that their primary objective of the study was to develop a model
for estimating TCDL grades based on ML algorithms which is essential for Tropical Cyclone disaster
prevention, risk mitigation, and decision making. We have taken the idea of TCDL(Tropical Cyclone Disaster
Loss) which eases our feature selection and the titles of the datasets which will be required in our project.
• Study on Typhoon Disaster Loss and Risk Prediction and Benefit Assessment of Disaster Prevention and
Mitigation[2] by Lin Yang, Chunrong Cao, Dehui Wu, Honghua Qiu, Minghui Lu, Ling Liu gives us an insight
about Typhoon Disaster Prediction and Risk Assessment. We have tried to implement the formula along with
our modification to the same which may give us the economic loss of the cyclones based on the historical
dataset. We may incorporate this calculation in this project.
• A study of deep learning algorithm usage in predicting building loss ratio due to typhoons: the case of
southern part of the Korean Peninsula[3] by Ji-Myong Kim, Junseo Bae, Manik Das Adhikari, Sang-Guk
Yum proposes a deep learning-based model for predicting building loss ratio due to typhoons. The model uses
a variety of features, including typhoon characteristics, building characteristics, and land use information, to
predict the probability of a building being damaged or destroyed by a typhoon. The model was trained and
evaluated on a dataset of typhoon damage data from the southern part of the Korean Peninsula. The results
showed that the proposed model outperforms traditional prediction models in terms of accuracy.
• Modelling tropical cyclone risks for present and future climate change scenarios using geospatial
techniques[4] by Muhammad Al-Amin Hoque, Stuart Phinn, Chris Roelfsema, Iraphne Childs remarks the
importance of considering climate change when developing risk models for tropical cyclones. The authors
propose a new approach to modelling cyclone risk that integrates geospatial data with climate change scenarios.
They test their approach in a case study of Sarankhola Upazila, Bangladesh. The results show that the new
approach can successfully identify areas at risk of cyclone inundation. The authors conclude that their approach
has the potential to be used to develop risk models for other coastal areas.

Department of Information Technology Page | 2

St. Thomas’ College of Engineering and Technology

1.5. Brief Discussion on Problem:

Tropical cyclones frequently impact the coastal states of India, causing substantial losses in human lives, livelihoods,
and infrastructure. Current cyclone prediction approaches primarily focus on track and intensity, often failing to
accurately represent the potential damage a cyclone can cause at the ground level.

Key issues include:

1. Track and Intensity vs. Actual Damage Potential: Predictions concentrate on cyclone track and intensity,
potentially overlooking the true damage potential.

2. Lack of Localized Impact Forecasts: Existing methods struggle to provide accurate ground-level impact
forecasts for specific localities, leading to inadequate preparedness.

3. Dependency on Location-Specific Factors: Cyclone damage is influenced by various location-specific factors

beyond wind speed, such as local infrastructure and population density.

4. Limited Actionable Intelligence: Vulnerable communities lack actionable intelligence to respond effectively to
the specific risks posed by incoming cyclones.

5. Challenges with Black Box Models: The use of traditional black box models impedes understanding and lessons
learned for model improvement.

Project Innovations:

The project aims to develop interpretable AI models correlating cyclone characteristics with casualties and economic
losses, addressing the aforementioned challenges. Key innovations include:

1. Integration of Meteorological Data with Local Statistics: Combining meteorological data with local
infrastructure and population statistics for a comprehensive impact understanding.

2. Explanations for Impact Predictions: Developing interpretable AI models providing explanations for impact
predictions to enhance trust and facilitate continuous improvement.

3. API-Based Prediction Delivery: Ensuring real-time predictions are integrated into existing early warning
systems through API-based delivery, enabling prompt and effective responses.

Expected Outcomes:
The project aims to enhance preparedness and response efforts, ultimately saving lives and protecting livelihoods in
vulnerable regions.

Department of Information Technology Page | 3

St. Thomas’ College of Engineering and Technology

2.0 Concept and Problem Analysis:

2.1 Target Users:

• Disaster management authorities.
• First responders.
• Local administrative bodies.

2.2 Hardware and Software Requirements:

Hardware:
• Server: Multi-core CPU , 16GB RAM, GPUs for accelerated computing
• Storage: Redundant fast disk storage with capacity for datasets and model files
• Backup: Tape drives for cold storage backup of data and models
• Workstations: High RAM, multicore workstations for development and analysis
Software:
• OS: Windows, Linux for servers & workstations
• ML frameworks: Python with Tensorflow, PyTorch, scikit-learn, NumPy
• Version control: Git/GitHub for code tracking
• Visualization: Matplotlib, Plotly, D3.js
• Notebooks: Jupyter for development & analysis
• Containers: Docker for deployment
• Databases: SQL, InfluxDB for efficient data storage & queries
Firmware:
• BIOS, BMC firmware on servers kept up to date for security and hardware optimizations
• Drivers: Latest GPU, NIC drivers for hardware accessibility & performance
Networking:
• Ethernet: Multiple 10Gbit and higher bandwidth low-latency Ethernet network
• Load balancers: For traffic distribution to model serving nodes
• Firewalls, VPN: For perimeter security and connecting remote locations

2.3. Estimation and Development Effort and Time:

As the project belongs to Organic Software Development project deals with a well-understood and size of the
development team is reasonably small. Assuming there be 216 lines of code in one single frame, then using the Basic
COCOMO Model we can estimate the effort and time required to develop the project.

Department of Information Technology Page | 4

St. Thomas’ College of Engineering and Technology

Thus, estimated effort to build the project is

15 modules of approx.
200 LOC each = 3000 LOC
Using COCOMO model,
Effort = 2.4 x (KLOC)^1.05 x PM
= 2.4 x (3)^1.05 x 1.4
= 11.88 Person Months
And estimated time to build the project is
Time = 2.5 x (Effort)^0.38
= 2.5 x (11.88)^0.38
= 7.2 Months.

2.4. Gantt Chart:

Fig. 1. Gantt Chart

Department of Information Technology Page | 5

St. Thomas’ College of Engineering and Technology

2.5. Methodology:
2.5.1. Work Flow Diagram:

LINEAR REGRESSION

RANDOM FOREST

Fig. 2. Work Flow Diagram

The Cyclone Casualty and Property Loss Prediction System follows a systematic workflow to predict the impact of
tropical cyclones using the Tropical Cyclone Disaster Loss (TCDL) framework. The process begins with data
acquisition, where relevant datasets are gathered from diverse sources such as meteorological agencies, disaster
authorities, and census/infrastructure bureaus. These datasets encompass information on past cyclones, their impacts,
demographics, and infrastructure parameters for affected areas. Subsequently, the acquired data undergoes thorough
processing to ensure its readiness for model building. This involves tasks like handling missing values, normalization,
and feature selection, with a focus on key variables identified by the TCDL framework, such as wind speed and
population exposure.

The next phase involves model training, where machine learning algorithms, including linear regression and random
forests, are employed to establish correlations between cyclone characteristics, geographical factors, and impact
metrics. The trained models are rigorously evaluated using test data, utilizing accuracy metrics such as Mean Absolute
Error and R-squared. This evaluation phase enables fine-tuning of the models to optimize their predictive performance.
In the inference stage, the best-performing model is employed to generate predictions of casualty and economic losses
based on incoming cyclone and location data. Overall, this systematic workflow ensures a comprehensive and data-
driven approach to cyclone impact prediction, leveraging the TCDL framework for enhanced accuracy and reliability in
anticipating the consequences of tropical cyclones.

Department of Information Technology Page | 6

St. Thomas’ College of Engineering and Technology

2.5.2. Linear Regression:

Linear regression is a type of supervised machine learning algorithm that computes the linear relationship between a
dependent variable and one or more independent features.

Why Linear Regression:

• Simplicity in Formulation and Training: Linear regression boasts an uncomplicated mathematical formulation
and model training process, making it easy to implement and interpret.

• Intuitive Predictions: The predicted losses in linear regression are expressed as a linear weighted combination
of factors, enhancing intuitive understanding of the model's output.

• Reduced Overfitting Risk: With few parameters compared to complex techniques like neural networks, linear
regression models are less prone to overfitting, especially when dealing with limited training data.

• Computational Efficiency: Closed-form mathematical solutions for optimizing linear models contribute to faster
training. Predictions involve basic matrix multiplications, enabling scalable deployment.

• Interpretable Baseline: Linear models provide an interpretable baseline, offering an intuitive lower bound for
the complexity required in modelling. This baseline helps identify factors with linear influence on losses,
serving as an explainable starting point.

Fig 3. Linear Regression

Linear Regression in our Project:

Build two separate linear regression models, one for predicting cyclone casualties and another for property losses. Use
the selected features as independent variables (X) and the target variable (casualties or property loss) as the dependent
variable (Y).
Evaluate the performance linear regression models using appropriate metrics such as Mean Absolute Error (MAE),
Mean Squared Error (MSE), and R-squared (R²). These metrics will help assess the model's accuracy and fit.

Department of Information Technology Page | 7

St. Thomas’ College of Engineering and Technology

2.5.3. Random Forest:

Random Forest is an ensemble technique capable of performing both regression and classification tasks with the use
of multiple decision trees and a technique called Bootstrap and Aggregation, commonly known as bagging.

Why random Forest:

• Nonlinear Relationship Handling: Random forests inherently model complex nonlinear relationships between
predictors (e.g., wind speed, rainfall, infrastructure damage) and loss outcomes without manual feature
engineering.
• Robust to Outliers: Ensemble averaging over decision trees makes random forests less sensitive to outliers in
cyclone disaster data, enhancing robustness.
• Avoids Overfitting: Random forests defend against overfitting by averaging and decorrelating multiple
decision trees, even with limited training data.
• Interpretability: Relative feature importance scores in random forests offer model transparency, valuable for
sensitive prediction tasks.
• Ease of Training: Random forests require minimal data pre-processing, can handle mixed data types, and
have few tuning parameters, facilitating almost off-the-shelf usage.

Fig.4. Random Forest

Random Forest in our Project:
• Random Forest is an ensemble of decision trees.
• In regression tasks, the leaf nodes of each tree hold the mean target variable values.
• To predict new data, it passes through all trees, aggregating the averages in leaf nodes.
• The final prediction is often calculated by averaging or taking the median of tree predictions.

Department of Information Technology Page | 8

St. Thomas’ College of Engineering and Technology

2.6. Use Case Diagram:

Fig.5. Use Case Diagram

Department of Information Technology Page | 9

St. Thomas’ College of Engineering and Technology

2.7. Context Free Diagram:

Fig.6. Context Free Diagram

Department of Information Technology Page | 10

St. Thomas’ College of Engineering and Technology

2.8. Level -1 DFD:

Fig.7. Level -1 DFD

Department of Information Technology Page | 11

St. Thomas’ College of Engineering and Technology

2.9. Entity Relationship Diagram:

Fig. 8. Entity Relationship Diagram

Department of Information Technology Page | 12

St. Thomas’ College of Engineering and Technology

3.0. Conclusion:
This project aims to develop an AI-based prediction system to forecast the potential human and economic impact of
impending cyclones. Advanced machine learning algorithms will be trained on historical disaster data to create models
that can provide localized, real-time casualty and loss forecasts.

• This project lays the foundation for a ground-breaking AI-powered system, using advanced machine learning
to forecast cyclone impacts.
• The interpretable modelling approach ensures actionable intelligence, empowering disaster management
authorities with targeted responses.
• From facilitating mass evacuations to aiding search and rescue efforts, the system addresses real-time
challenges.
• Recognizing the urgency posed by climate change, the solution harnesses leading-edge data science to enhance
accuracy in impact forecasting.
• The integration with early warning systems empowers communities, fostering disaster resilience.
• Future enhancements, including increased accuracy and expanded coverage, aim to unlock the full lifesaving
potential of AI in cyclone preparedness.
• In conclusion, this project signifies a pivotal step towards an AI-powered, location-aware impact prediction
system, poised to revolutionize disaster management for various natural hazards.

FUTURE SCOPE:

The system can be enhanced by:

• Incorporating more datasets to improve accuracy.
• Testing additional algorithms like neural networks.
• Developing mobile/web interfaces for easy access to predictions.

Department of Information Technology Page | 13

St. Thomas’ College of Engineering and Technology

Annexure:

References:

[1] https://www.mdpi.com/2071-1050/15/16/12261
[2] https://www.sciencedirect.com/science/article/pii/S2225603219300074
[3]https://www.frontiersin.org/articles/10.3389/feart.2023.1136346/full
[4]https://www.tandfonline.com/doi/full/10.1080/17538947.2017.1320595

Department of Information Technology Page | 14