KABARAK UNVERSITY

SCHOOL OF SCIENCE ENGINEERING AND TECHNOLOGY

DEPARTMENT OF COMPUTER SCIENCE

COMP 323

TITLE: Weather Forecasting Using Machine Learning

SUBMITTED BY

NAME: HENRY MUGO GACHUKI

CS/MG/0797/05/18

NAME: BRIAN KIRWA

CS/M/1412/09/16

NAME: BRIAN CHERUIYOT

INTE/M/0386/05/16

SUPERVISOR: MR MARTIN MUTAI

SUBMITTED ON: JULY, 2022

This research proposal is submitted to Kabarak University Department of Computer

Science in partial fulfilment of the requirement for the award of science degree of
computer science.

i
 STATEMENT OF ORIGINALITY

I declare that this is my original work and to the best of my knowledge it has never been
presented in this University or any other Institution of higher learning.

HENRY MUGO SIGNATURE……………………………………….

REG NO: CS/MG/0797/05/18

BRIAN KIRWA SIGNATURE……………………………………….

REG NO: CS/M/1412/09/16

ii
DECLARATION BY SUPERVISOR

This thesis is the candidates original work and has been prepared with my guidance and
submitted with my approval as the official university project.

………………………………………………………………………………………………..

Mr. MARTIN MUTAI

Signature…………………………….

Date………………………………….

Department of computer science and information technology

iii
iv
 ACKNOWLEDGEMENT

We would like to thank the almighty God for granting us the strength, knowledge and ability
to successfully complete this project.

Secondly, we gratefully acknowledge the kind guidance and expertise of Mr. Mutai
throughout this project. Much appreciation for impacting us with the necessary knowledge
and skills.

Finally, we thank the entire Kabarak fraternity for its extensive opportunity to enable us to
pursue our studies and also equipping us with the necessary skills applicable in this project.

v
 ABSTRACT
Weather is one of the most effective environmental constraints in every phase of our lives on
the earth. We need to predict the weather such as temperature, rainfall and humidity to
protect ourself from abnormal conditions. Weather conducts a completely critical function in
many key production sectors. Weather changes with high charging these days, which is why
old weather forecasts are getting closer and less powerful and continue to be annoying.
Predictions affect the county’s financial system and people’s lives. A system of information
and statistics analysis algorithms has been used that includes a wooded area used for weather
forecasting. The weather is one of the highest natural barriers in all parts of our lives in the
world, we need to look at the weather including temperature, rain, humidity and other
protection. Earth's climate will change over a long period of time and also what kind of
impact it will have on the lives of future generations. Our predictive nature of end-of-life
climates offers an excellent desire to provide information as a way to allow stadium insurers
to make an informed wish for the future of the world. Our approach greatly enhances the
model in a positive way to govern the state of staff inconsistencies and inequalities and
performs its function of accurately predicting the weather. This paper explores three machine
learning models for weather prediction namely Support Vector Machine (SVM), Artificial
Neural Network (ANN) and a Time Series based Recurrent Neural Network (RNN). RNN
using time series along with a linear SVC and a five-layered neural network is used to predict
the weather. The results of these models are analysed and compared on the basis of Root
Mean Squared Error between the predicted and actual values. Hence Time Series based RNN
does the best job of predicting weather.

vi
TABLE OF CONTENTS

vii
List Of Figures

viii
 CHAPTER ONE:

Introduction
1.1: Background
Weather forecasting is the application of science and technology to predict the
conditions of the atmosphere for a given location and time. People have attempted to
predict the weather informally for millennia and formally since the 19th century.
Weather forecasts are made by collecting quantitative data about the current state of
the atmosphere, land, and ocean and using meteorology to project how the atmosphere
will change at a given place. Ancient weather forecasting methods usually relied on
observed patterns of events, also termed pattern recognition. For example, it was
observed that if the sunset was particularly red, the following day often brought fair
weather. This experience accumulated over the generations to produce weather lore.
However, not all of these predictions prove reliable, and many of them have since
been found not to stand up to rigorous statistical testing.

It was not until the invention of the electric telegraph in 1835 that the modern age of
weather forecasting began. The electric telegraph network expanded, allowing for the
more rapid dissemination of warnings, a national observational network was
developed, which could then be used to provide synoptic analyses. To convey
accurate information, it soon became necessary to have a standard vocabulary
describing clouds which was achieved. On the 20th century, advances in the
understanding of atmospheric physics led to the foundation of modern numerical
weather prediction.

1.2 Problem Statement

Weather data is unstable in nature which makes forecasting weather with current
measurements less accurate. However, with the help of machine learning techniques
weather predictions can be achieved more accurately. This paper proposes the use of
Linear Regression, Regression Decision Tree, Binary Logistic Regression, Decision
Tree Classification, Principal Component Analysis and Clustering techniques of
machine learning to forecast weather achieving higher accuracy.

9
1.3 Main Objectives
The aim of this project is to develop a machine learning algorithm that will be an
effective prediction agent weather model.

1.3.1 Specific Objectives

The specific objectives of this project are:
I. To analyse the existing system solutions put in place to predict weather.
II. To evaluate the challenges in existing algorithms to predict weather.
III. To design, develop and test a weather forecasting algorithm that will assist in
weather forecasting.

1.3.2 Research Questions

I. What are the existing solutions?
II. What are the challenges in the existing algorithms?
III. What is the proposed solution?

1.4 Justification
Weather forecasting is challenging due to hyper variability of climate observations
recorded in a day. According to traditional approach of predicting weather,
simulations are performed manually using numeric equations of thermodynamics and
fluid dynamics. However, uncertain weather conditions restrict the estimation of
weather prediction and makes it inaccurate following traditional methodologies. On
contrary, machine learning provides supervised and unsupervised learning techniques
to forecast weather with minimal error. It doesn’t demand in depth understanding of
climate affecting variables and atmospheric theories. Machine learning provides
robust and highly significant results based on current observations to forecast the
weather conditions for future.

10
1.5 Scope of the project
We can’t able to use linear regression when it comes to huge amount of data set and
as its doesn’t give accurate result. So, for predicting huge volume of dataset we can
develop a neural network system for more better results and accurate prediction of the
weather forecasting. Also, we connect analysing process to IOT technology because
without data we cannot perform analysis and prediction because IOT is major source
of data. So IOT will generate data from devices which helps to take initiative to
improve decision making.

1.6 Limitations of the project

Machine learning in weather forecasting is not a fully developed ecosystem. It is con-
tinuously evolving as more countries and organizations invest in building better data
governance and information exchange systems globally for seamless weather predic-
tion capabilities. There are limitations for present MI models such as limited data
availability for certain weather conditions, gated information highways set up by cer-
tain countries, differences in technology and hardware standards used by weather sys-
tems across different countries and organizations.
However, ML will be a key pillar in enabling mankind to predict and handle nature’s
fiercest avatars and create widespread socio-economic safety frameworks globally
against weather uncertainties. 

CHAPTER 2
2.0 LITERATURE REVIEW

11
2.1 Introduction

In the daily lives of human beings, their activities are usually influenced and altered
by changing weather patterns and therefore a quite accurate system that tells how the
weather will be is necessary. Weather is a continuous, dynamic chaotic process and
data-intensive which makes weather forecasting one of the challenging difficulties in
the world even for meteorologist. Development of a more accurate weather
forecasting system will be more effective in the daily lives of human. Presence of
such a system will help save lives in that, it predicts the occurrences of weather-
related calamities like floods, tsunami and even drought. Nowadays it’s not surprising
to hear that some people are homeless and others have died for reasons such as
flooding which was never foretold. Changes in these weather patterns are more
frequent as a result of human activities which even cause global warming making it
necessary for such an application. Existence of such a system will also help save
money, in the essence that organizations such as government and other bodies can
know what danger can occur in future and will be caused by weather. These
organizations will be able to work earlier before its occurrence which will be much
cheaper than on its occurrence. Such a system will also help persons like farmers who
rely on weather and also fishermen who may encounter different weather patterns in
the seas and ocean. Existence of such a system can also help in finding other ways of
doing things like production of electricity by wind. Which is a cheaper and a safer
way. Either accurate and timely weather forecasting is a great challenge. Predication
of a weather model involves combination of many computers, collection of data,
observation, design and trends. Use of these models can bring an accurate system.

2.2 General overview

Weather forecasting provides critical stimulation about future weather. There are
various techniques that include in weather forecasting, from easy sighting of the sky
to highly complicated computerized mathematical stimulation. To form associate
correct prediction is one in every of the key challenges facing meteorologist
everywhere around the globe. Back in the days, weather prediction has been one in
every of the foremost attention-seeking and interesting aspect. Scientists have tried to
forecast meteoric characteristics employing arrange of ways, a number of these ways
being a lot of correct than others. Data recorded by a Weather Station at the

12
outstanding scientific discipline centres are used for the analysis. Some techniques are
capable of handling non - stationary statistic knowledge, uses variable’s lag values
combining with previous and present random residual item to reveal the variation of
variables. That is, at the same time it takes into account all to regressive with
distinction variables. It improves the static multivariate analysis and thus it's
additional wide use to predict forecasting. Meteorological data mining is a form of
Data mining concerned with discovering hidden designs inside largely existing
meteorological data, so that the information retrieved can be converted into usable
knowledge. Weather forecasting is a vital issue within the field of science on the
globe. Many factors contribute considerably to extend the forecasting accuracy one in
all that the development of applied statistic always for enhancing the scope and
accuracy of model prediction. Several efforts are created by researchers to spot the
simplest precise forecasting models. The combination of linear and nonlinear models
is one in every of the foremost well-liked and was wide used in forecasting models for
the forecasting accuracy.

2.3 Literature review based on objective one

Scientist all over the world have been working on an efficient system that will have
accurate data within a short span of time in order to help save people in case of a bad
occurrence. But many of these systems do not give that accurate information hence
occurrence of calamities.

Example includes the Case Study: Abadeh station, Iran. Mahsin et al. (2012) used
Box-Jenkins methodology to form seasonal ARIMA model for monthly weather
information taken for Dhaka station, Bangladesh, for the significant amount from
1981-2010. In their paper, ARIMA (0, 0, 1) (0, 1, 1) model was found suitable and
also the model is working for forecasting the monthly climate prediction. Marmara
University, Vocational School of Technical Sciences, Turkey(2010) comparative
study of statistical and neuro-fuzzy network models for forecasting the weather of
Göztepe, İstanbul, Turkey Adaptive Network Based Fuzzy Inference System and
Auto Regressive Moving Average models have been applied that ensure the
efficiency of ARIMA and ANFIS techniques. Different models using a different
training and test data set have been tested and the criteria of performance evaluation

13
which are calculated for estimating and then comparing the performances of ARIMA
and ANFIS models for forecasting has mentioned the clear review.

Accurate systems can be used to predict impact of weather if it occurs. Many systems
developed only predict patterns but are not reliable to predict the impact of its
occurrence. This aspect should be considered in such a system.

2.4 Literature review based on objective two

The length or duration of an impact lasts in case it occurs is one of the gaps of that
should be bridged in development of a weather forecasting system. Determination of
this length will help organizations in charge to make appropriate plans and avoid lose
of life if such an occurrence happens.
In the author Kuo and Sun, (1993) was used to Associate in having intervention
model for average of 10 days stream flow forecast and synthesis that was investigated
by to affect the extraordinary phenomena caused by typhoons and different serious
irregularities of the weather of the Tan Shui geographical area in Taiwan. In Chie wet
al, (1993) conducted a comparison of six rainfall-runoff modelling approaches to
daily, monthly and annual flows in eight tolerant catchments.
They concluded that time-series approaches with agreement adequate estimates of
monthly and annual yields within the water capitals of the catchments. Determination
of duration also helps personnel like farmers to prepare in advance which makes
production of food crops continuous.

2.5 Literature review based on objective three

Example: Langu, (1993) is another approach which used statistical analysis to observe
changes in weather and runoff patterns to go and look for important changes inside
the parts of variety of weather statistic. Box and Jenkins (1994), in early 1970's,
pioneered in developing methodologies for statistic indicating within the univariate
case often known to Univariate Box-Jenkins (UBJ) ARIMA modelling in this
approach of the author.
Another gap that should be bridged is accuracy of the information given. Most of
daily lives depend on the information given by such systems.

14
2.6 Design framework
Many systems for weather forecasting have been created but the main outstanding
problem is there’s no accurate system. Since this is a critical issue since its applied in
flying of planes and even on water vessels which deals with human lives it should be
95% or even 100% accurate. But since factors that changes weather is not determined
by human activities it should give the better as it can.

15
CHAPTER 3

METHODOLOGY

3.1 Introduction
3.2 Software Development Methodology
3.2 System Design
3.2.1 Use Case Diagram
3.3.2 Class Diagram
3.3.3 Entity Relationship Diagram
3.3.4 Database Schema

16
CHAPTER 4

SYSTEM ANALYSIS AND DESIGN

4.1 Introduction
4.2 System Implementation User Interfaces

17
Chapter 5: System Implementation and Testing
Discussion

Recommendation

5.3 Conclusion

18
19