New provide essential information and could lower the number

New
Rainfall Prediction using SVM and DFO algorithms 

Mohamed
Bhailat ?January 2018 

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!


order now

 

Introduction 

Rainfall prediction
has become one of the most scientific and technologically challenging issues in
the world today. Rain prediction is also very imperative within weather
forecast. The rain prediction model has attracted the attention of
governments, Industries such as risk management to name but a few.

 

Why
are rainfall predictions of such importance? 

The
climate can affect human activities such as agricultural production,
construction, power generation and Tourism. Therefore,
having an appropriate approach for precipitation forecasting makes it possible
to take the appropriate measures depending of the field;  

 

An accurate prediction of precipitation will facilitate for the
necessary supervision of agriculture activities. In this case it is necessary
to classify weather, such as for the next day or next month to decide time of
planting. It is also essential for the planning and management of
water resources around the world, and more in semi-arid area as south of Morocco. 

 

A good prediction of rainfall can be very important for the
safety of a population, such as in the case of flash flooding (intense
rainfall). A good flood warning system will provide essential information and
could lower the number of casualties in the case of a natural disasters. 

 

Statement
of Problem 

There
are two main approaches taken when predict rainfall: 

 

-The
dynamical method, which is based on satellite 

-The
empirical approach which is based on analysis of historical data.  

 

The
project will focus on the second approach. 

To
achieve predictions there are a number of methods ranging from simple data
mining i.e. Naïve Bayes or KNN, to those that use much more complex techniques
such as: 


Artificial Intelligence(AI) 


Artificial Neural Networks(ANN) 


Deep Learning 1 


NN using hybrid Particle Swarm Optimisation and Genetic Algorithm (HPSOGA) 2. 

 

A
paper of interest currently in use, Particle Swarm Optimisation and Support
vector machine(PSO-SVM) 3. 

 

PSO
is an evolutionary computation technique. Which can deal with the continuous
optimisation problem. PSO was used to achieve the best parameters for
SVM. 

SVM
is algorithm used in many machine learning problems, it takes two parameters C
and g (C is included {2^-5; 2^15}, g is included {2^-15; 2^3}) which are the
parameters which they used to tune and achieve the best predictions in the
model. 

 

Objectives 

The
goal of the project will be to use DFO (Dispersive Flies Optimisation) over PSO
and compare my results with other techniques such as PSO-SVM or HPSOGA which
gave the best results for rainfall prediction. 

DFO
introduced by Mohammad Majid Al-Rifaie is a population-based stochastic
algorithm that stimulates the collective behaviour of flies searching for
food. 

DFO
was chosen because it has less parameters than PSO, which makes it simpler to
implement, it also yielded better results in many research. 

 

Plan
of Action 

This
section presents the plan for obtaining the objectives discussed in the
previous section.  

•  
Implementation
using Octave/Matlab or Java or Python 

•  
RapidMiner
software to check my results 

•  
Get
dataset 

•  
Implement
a simple SVM  using a small dataset 

•  
add
implementation of DFO 

•  
datapreparation 

•  
use
real data once the implementation work for small dataset 

•  
implement
other methods 

•  
compare
results of my method against others 

 

 

An
email was sent to the author who wrote paper 3 which is of interest, to get
the same data (one year ground based meteorological data from Nanjing Station (ID:
58238), which contains as attributes: atmospheric pressure, sea level, wind
direction, wind speed, relative humidity and precipitation. 

If
data is not forwarded in time, the Texas data set will be used, and both methods
will be implemented to enable the compareson. 

 

Management
Plan 

This
section presents my schedule for completing the proposed research. This
research culminates in a formal report using LaTeX, which will be completed by
june 1, 2018. To reach this goal, I will follow the schedule presented in
Figure 1.  

 

New
Rainfall Prediction using SVM and DFO algorithms 

Mohamed
Bhailat ?January 2018 

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!


order now

 

Introduction 

Rainfall prediction
has become one of the most scientific and technologically challenging issues in
the world today. Rain prediction is also very imperative within weather
forecast. The rain prediction model has attracted the attention of
governments, Industries such as risk management to name but a few.

 

Why
are rainfall predictions of such importance? 

The
climate can affect human activities such as agricultural production,
construction, power generation and Tourism. Therefore,
having an appropriate approach for precipitation forecasting makes it possible
to take the appropriate measures depending of the field;  

 

An accurate prediction of precipitation will facilitate for the
necessary supervision of agriculture activities. In this case it is necessary
to classify weather, such as for the next day or next month to decide time of
planting. It is also essential for the planning and management of
water resources around the world, and more in semi-arid area as south of Morocco. 

 

A good prediction of rainfall can be very important for the
safety of a population, such as in the case of flash flooding (intense
rainfall). A good flood warning system will provide essential information and
could lower the number of casualties in the case of a natural disasters. 

 

Statement
of Problem 

There
are two main approaches taken when predict rainfall: 

 

-The
dynamical method, which is based on satellite 

-The
empirical approach which is based on analysis of historical data.  

 

The
project will focus on the second approach. 

To
achieve predictions there are a number of methods ranging from simple data
mining i.e. Naïve Bayes or KNN, to those that use much more complex techniques
such as: 


Artificial Intelligence(AI) 


Artificial Neural Networks(ANN) 


Deep Learning 1 


NN using hybrid Particle Swarm Optimisation and Genetic Algorithm (HPSOGA) 2. 

 

A
paper of interest currently in use, Particle Swarm Optimisation and Support
vector machine(PSO-SVM) 3. 

 

PSO
is an evolutionary computation technique. Which can deal with the continuous
optimisation problem. PSO was used to achieve the best parameters for
SVM. 

SVM
is algorithm used in many machine learning problems, it takes two parameters C
and g (C is included {2^-5; 2^15}, g is included {2^-15; 2^3}) which are the
parameters which they used to tune and achieve the best predictions in the
model. 

 

Objectives 

The
goal of the project will be to use DFO (Dispersive Flies Optimisation) over PSO
and compare my results with other techniques such as PSO-SVM or HPSOGA which
gave the best results for rainfall prediction. 

DFO
introduced by Mohammad Majid Al-Rifaie is a population-based stochastic
algorithm that stimulates the collective behaviour of flies searching for
food. 

DFO
was chosen because it has less parameters than PSO, which makes it simpler to
implement, it also yielded better results in many research. 

 

Plan
of Action 

This
section presents the plan for obtaining the objectives discussed in the
previous section.  

•  
Implementation
using Octave/Matlab or Java or Python 

•  
RapidMiner
software to check my results 

•  
Get
dataset 

•  
Implement
a simple SVM  using a small dataset 

•  
add
implementation of DFO 

•  
datapreparation 

•  
use
real data once the implementation work for small dataset 

•  
implement
other methods 

•  
compare
results of my method against others 

 

 

An
email was sent to the author who wrote paper 3 which is of interest, to get
the same data (one year ground based meteorological data from Nanjing Station (ID:
58238), which contains as attributes: atmospheric pressure, sea level, wind
direction, wind speed, relative humidity and precipitation. 

If
data is not forwarded in time, the Texas data set will be used, and both methods
will be implemented to enable the compareson. 

 

Management
Plan 

This
section presents my schedule for completing the proposed research. This
research culminates in a formal report using LaTeX, which will be completed by
june 1, 2018. To reach this goal, I will follow the schedule presented in
Figure 1.  

 

x

Hi!
I'm Elaine!

Would you like to get a custom essay? How about receiving a customized one?

Check it out