Forecasting Weather Status Using Advanced Machine Learning

Algorithm

Karuppusamy S., Soundarraj S., Vasanth P. and Vijayasankar T.

Department of Computer Science and Engineering, Nandha Engineering College, Erode, Tamil Nadu, India

Keywords: Weather Forecasting, Machine Learning, XG Boost Algorithm, Meteorological Data, Temperature Prediction,

Climate Change, Agricultural Risk Management, Flask Framework, Extreme Gradient Boosting.

Abstract: Weather forecasting in agriculture is, indeed, rather difficult because the whole area is so dynamic and

variable. Conventional statistical approaches usually cannot provide excellent precision, and this makes it a

challenge for farmers to be and plan effective. The project concentrates on an accurate temperature prediction

mechanism, with the application of machine learning methods, which includes the analysis of past as well as

current-day weather data to increase the reliability of predictions. Despite advances in forecasting techniques,

challenges remain, especially, in the areas of improving the accuracy of models, validating climate forecasts

in agricultural risk management, and their effect on crop diseases seasonally. As temperature and rainfall are

the two most crucial factors influencing plant health and production, the only advanced predictive system

available may provide farmers with insightful decisions in preventing losses. This project, therefore, aims at

an integrated assessment of weather forecasting techniques to improve agricultural planning strategies and

climate adaptation via data-driven approaches.

1 INTRODUCTION

Weather forecasting is key in various sectors, which

include agriculture, transport, disaster management,

and urban planning. To derive actions necessary to

avert possible risks once the availability of the

forecast is known is paramount. Some traditional

forecasting methods may not respond satisfactorily to

these dynamic and nonlinear changes that take place

within the atmosphere. This has opened avenues for

development in machine learning algorithms and

better scope with their promise of increased accuracy

i n weather predictions. Machine learning

approaches-XG Boost, in particular now have made

considerable advances successfully analyzing large

datasets and offers enormous improvement in

predictive accuracy. The proposed work aims to

design an extreme gradient boost for the forecasting

system which is a combination of historical and real-

time meteorological data to make accurate forecast

predictions. The project aims to provide farmers,

meteorologists, and policymakers with timely

information and decisions to allow proper planning

and avoid losses due to unusual weather. Through the

preprocessing of data techniques and feature

engineering, and integration of complex machine

learning algorithms, the presented system guarantees

higher prediction accuracy and timely alerts of

extreme weather phenomena. Finally, the underlying

principles make provisions for better climate

adaptation strategies in improved risk management

across different sectors.

2 RELATED WORKS

Machine learning and AI weather prediction models

have all but outrighted the prediction capabilities

using deep learning, numerical weather prediction,

and data-driven models. There have been several

studies addressing the advanced use of random

forests, CNNs, LSTMs, and graph-based neural

networks with the aim of improving weather

forecasting models (TL Yu, et.al., 2024). The shift

toward the data-driven forecasting and the

acknowledgment by researchers about the

capabilities an AI model might have in enhancing

both global and regional weather predictions (Ben-

Bouallegue, et.al., 2023). AI approaches, Four Cast

Net and Graph Cast, much faster in boosting scientific

progress in weather modeling (Anandkumar, A.

S., K., S., S., P., V. and T., V.

Forecasting Weather Status Using Advanced Machine Learning Algorithm.

DOI: 10.5220/0013929800004919

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 5, pages

351-356

ISBN: 978-989-758-777-1

351

(2024)). Also, these studies deem that the

involvement of deep learning models partly

incorporated with physical processes using physical

constraints ensures more accurate predictions (de

Bezenac, et.al., 2020). Res Net-based models and

deep convolutional networks provided promising

results for medium-range forecasting (Rasp, S., &

Thuerey, N. (2021)), and also adaptive Fourier neural

operators have been very well used for high-

resolution forecasts (Pathak, J, et.al., 2022).

Moreover, graph neural networks have efficiently

predicted weather patterns showing good spatial

awareness (Sønderby, C. K., et.al., 2020). In addition,

the AI-powered precipitation forecasting models such

as Met Net have significantly improved short-term

weather forecasts Zhang, J., et.al., 2021.

The integration of convolution neural networks in

satellite image analyses has played an important role

in the storm detection and severe weather prediction

Molchanov,et.al., 2021. Further, other studies

explored the use of IoT-sensor data fusion with AI

techniques for real-time weather monitoring (Sharma,

R., et.al.(2022)). Thus, big data analytics

incorporated into cloud-based prediction systems

have greatly enhanced AI's predictive capabilities

(Weyn, J. A., et.al.(2020)). In addition, methods of

ensemble learning and data assimilation are already

being investigated so as to produce the optimal

machine learning model for weather forecasting

(Kashinath, K.,et.al., (2021) ) Innovative

developments in meteorology to streamline the

accuracy of prediction include physics informed deep

learning method (Evensen, G., & Monsen, S. M.

(2021). ). All of these undertakings further exemplify

the thrust of artificial intelligence, deep learning, and

big data analytics into modern weather forecasting,

thereby heralding a more reliable and intelligent

predictive system.

3 DATASET COLLECTION AND

PRE-PROCESSING

3.1 Dataset Collection

To inform the forecasts, this project intends to use the

most diversely sourced high-quality data from widely

reputable sources TL Yu, et.al.,2024. The data

sources include: satellite observations, ground- based

stations, remotely sensed technologies, Internet of

Things enabled sensors, and historical weather

records. Satellite observations have a wide variety of

available data about the atmosphere, including

temperature, humidity, cloud amount, wind speed,

and precipitation level. (Ben-Bouallegue , et.al.,2023)

This forms the basis for creating different long-term

weather patterns and events of severe weather on a

broad scale (Anandkumar, A. (2024)). Ground- based

weather stations supply meteorological data such as

real-time information on atmospheric pressure, wind

direction, temperature variations, and precipitation.

(Anandkumar, A. (2024).)This plays a major role in

solving satellite readings through providing further

precision in forecasting locally.(de Bezenac,et.al.,

(2024)) Remote sensing technologies add the frenetic

capability of latest technologies, for instance, Doppler

RADAR and LIDAR in observing clouds-storm

intensity- wind currents, which is a further push

towards intrusive short-term weather

predictions.(Rasp, S., & Thuerey, N. (2021) The

technologies allow for extreme monitoring of

phenomena like hurricanes and

thunderstorms.(Weyn, J. A.,et.al.(2021)) IoT-enabled

weather sensors, which are located locally, collect

real-time meteorological data that enhance

microclimate analysis and short-term

forecasting.(Pathak, J. ,et.al.(2021)) Long term

historical weather records gathered together,

including NOAA, Kaggle, and meteorological

agencies basically form the background for training

machine learning models in detecting trends and

forecasting future events.(Sønderby, et.al.(2020))

Weather APIs, such as Open Weather Map, Weather

Stack, and Climacell, provide such Public API across

different parts of the globe and ease the task of the

meteorologist taking into account accuracy on

forecasts.

3.2 Data Pre-Processing

Raw meteorological data are often marked in-

completeness, inconsistency, and noise, and several

preprocessing phases are performed in order to enable

high-quality machine learning model input that

effectively cleans, normalizes, and structurally

organizes the dataset (Ben-Bouallegue, et.al., 2023).

Dealing with missing data: Value missingness in a

weather dataset can occur with sensor-related failure

or incomplete transmissions. To handle missingness,

the most often used statistical imputation strategies

include but are not limited to mean, median, mode

replacement, or interpolation (de Bezenac,et.al.,

2020). Noise reduction: Due to environmental

conditions, sensor readings may be affected, resulting

in fluctuations in the recorded values. Moving

averages, median filters, and outlier removal

algorithms are helpful for smoothing the data and

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

352

getting rid of noise. Feature scaling and

normalization:

Different weather parameters such as

temperature, wind speed, and humidity exist on

different scales. (Pathak, J.,et.al., 2022) To resolve

that problem, min-max normalization and

standardization are applied to bring all features to a

common scale so that the model interprets them

correctly Zhang, J., et.al., 2021. Anomaly detection

and correction: Spikes or big drops in data can

generally indicate faulty sensors or extreme weather

conditions. Anomaly detection techniques based on

machine learning, such as Z-score and interquartile

range (IQR) methods, can be employed to identify

and remedy these anomalous data Molchanov,et.al.,

2021. Feature engineering: These are other

interesting features that can be derived to improve

prediction accuracy.

4 PROPOSED METHODOLOGY

Advanced machine learning and deep learning

algorithms were tackled in the proposed scheme in

order to classify the kinds of weather being reported

and future prediction of weather. It comprises

modeling that takes multiple phases, including object

detection, time-series forecasting, and others, for the

precise and reliable prediction of weather conditions.

Figure 1: XG Boost algorithm.

The object detection algorithms analyze satellite

imagery and video feed to identify numerous patterns

of various weather conditions. The selection of the

detection model depends on the complexity of the

data and the processing speed required. Real-time

detection is performed using YOLO (You Only Look

Once) with a focus on speed, while higher precision

and accuracy in detecting satellite images involve the

usage of Faster R-CNN algorithms. For scenarios with

multiple kinds of weather patterns, SSD (Single Shot

Multibook Detector) is used, while Mask R-CNN

allows for pixel-wise segmentation to determine the

size and shape of cloud formations and storm

structures. This involves time series forecasting with

models such as ARIMA and LSTM, which are perfect

for recognizing trends concerning storm movement,

precipitation changes, variation in the atmospheric

pressure, and cyclonic activities. Further, this model

generates accuracy in the predictions by learning from

both real-time and historic meteorological data with

an over-expectation factor of improvement in the

short-term weather forecast. In order to enhance the

reliability in making predictions and forecasts, the

system combines the Artificial Intelligence and

Forecasting Weather Status Using Advanced Machine Learning Algorithm

353

machine learning techniques with conventional NWP

(Numerical Weather Prediction) models. The

approaches of Support Vector Machines (SVMs) and

Neural Networks are capable of capturing the data-

dependent forecast in passing their short-term

weather prediction ahead along with the detection

systems that identify the onset of any severe event like

tornadoes or hurricanes. To perform a reliable and

rigorous evaluation with numerous metrics, this

project has employed precision, recall, and

Intersection over Union. Precision is the proportion

between every weather event classified as positive

and the total number of weather events, thus

minimizing false positives. Recall deals with how

many relevant meteorological phenomena were

detected by the model, minimizing ratio of false

negatives. Intersection over Union measures the ratio

of the predicted weather conditions over the actual

data for the clear localization of detected events.

Figure 1 shows XG Boost algorithm. Such methods

are expected to inject more reliability into forecasts

whereby automation of the alert system for weather

hazards expects to issue such alerts in timely fashions

to assist sound decision-making. Figure 2 shows the

Flow diagram.

Figure 2: Flow Diagram.

5 EXPERIMENTAL RESULT

Experimental analysis of the XG Boost-based model

for weather prediction confirmed weather predictions

high accuracy, efficiency, and reliability through the

use of historical and real-time meteorological data.

The data was collected from NOAA, Open Weather

Map API, and IoT sensors. Figure 3 shows

Temperature result. The data was also subjected to

feature engineering, anomaly detection, and

normalization, thereby constituting 80% training and

20% testing. The model was evaluated with major

performance metrics, i.e., RMSE, MAE, and R² score,

and was seen to outperform traditional models like

ARIMA, LSTM, and Numerical Weather Prediction

(NWP) methods. XG Boost model came out with an

RMSE of 1.25°C, MAE 0.85°C, and R² score of 0.92;

it required about 2.5 seconds in actual computation,

making it decidedly more efficient than conventional

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

354

forecasting techniques. Figure 4 shows Humidity

result. In addition, some deep learning- based

algorithms (YOLO, Faster R-CNN, Mask R-CNN)

were integrated to analyze satellite images and radar

data thereby increasing from 87% accuracy in issuing

early warnings for cyclones, heavy rainfall, and

severe weather events. The model was subsequently

deployed using a Flask API and operated as a real-

time forecast tool via a web-based dashboard. The

applications received lots of praise from

meteorologists, farmers, and disaster management

teams, which deemed them useful for short-term

weather forecasts, as well as agricultural planning and

disaster preparedness. Future improvements entail

implementing ensemble learning, cloud computing,

and hybrid AI with numerical weather prediction to

achieve further reduction in forecast uncertainty and

improve accuracy and scalability. Results have shown

machine learning to herald a new horizon for

adaptive, data-driven, and responsive weather

forecasts. Figure 5 shows wind result.

Figure 3: Temperature result.

Figure 4: Humidity result.

Figure 5: Wind result.

6 CONCLUSIONS

Hence, being very good in demonstrating the power of

the machine-learning forecasting using XGBoost, this

project integrates meteorological data of various

channels, includes more complex data preprocessing

steps, and extremely accurate predictions on major

weather parameters like temperature, humidity, wind

speed, and precipitation. Contributions toward the

analysis of satellite imagery in extreme weather

events will also allow those algorithms to promote

further research toward breaching that wall. Future

development includes merging deep learning to cloud

spectrum and ensemble learning in improving the

accuracy of forecasting and real- time adaptability.

In the planned future improvements of the project,

model accuracy and efficiency would be enhanced by

hybrid AI, deep learning algorithms, and big data

analytics. One of such improvements could be in

developing global weather predictions where having

real-time cloud-based processing could enhance its

scalability and efficiency. This includes investigating

the use of other methods of ensemble learning models

and automation to stitch together several forecasting

techniques under one roof to enhance reliability. The

system can be scaled to predict extreme weather events

with higher accuracy through improved satellite

image analysis and machine learning algorithms.

Further, combining IoT based smart weather stations

and predictive analytics can enable hyperlocal

forecast benefits for agriculture, transportation, and

disaster management.

Forecasting Weather Status Using Advanced Machine Learning Algorithm

355

REFERENCES

Anandkumar, A. (2024). Anima Anandkumar is

Accelerating Scientific Discovery with AI. TIME.

Ben-Bouallegue, Z., Clare, M. C. A., Magnusson, L., et al.

(2023). The rise of data- driven weather forecasting.

de Bezenac, E., Pajot, A., & Gallinari, P. (2020). Deep

learning for physical processes: Incorporating prior

scientific knowledge.

Evensen, G., & Monsen, S. M. (2021). Review of data

assimilation in AI-based weather prediction. Weather

and Climate Dynamics, 2, 317-333.

Google DeepMind (2024). Google DeepMind predicts

weather more accurately than leading system. The

Guardian.

Journal of Statistical Mechanics: Theory and Experiment,

2020(12), 124009.

Kashinath, K., Mustafa, M., & Albert, A. (2021). Physics-

informed deep learning approaches for improving

weather prediction. Proceedings of the IEEE, 109(4),

555-571.

Molchanov, S., Kitaygorodskaya, Y., & Mishin, D. (2021).

Examined the fusion of AI with IoT sensor data to make

predictions in real- time about the weather. Sensors,

21(14), 4852.

Pathak, J., Subramanian, S., Harrington, P., et al. (2022).

FourCastNet: A global data-driven high-resolution

weather model using adaptive Fourier neural operators

Keisler, R. (2022). Forecasting global weather with

graph neural networks.

Photovoltaic Power Generation Forecasting Based on

Random Forest Algorithm TL Yu, H Liang, L Jian 2024

8th International Conference on Green Energy and

Applications. [2024] ieee.

Rasp, S., & Thuerey, N. (2021). Data-driven medium-range

weather prediction with a Resnet pretrained on climate

simulations: A new model for weather forecasting.

Journal of Advances in Modeling Earth Systems.

Sharma, R., Gupta, A., & Patel, N. (2022). Suggested

cloud-based big data analytics as a mechanism for AI-

Driven weather prediction. Journal of Big Data, 9(1),

45.

Sønderby, C. K., Wang, S., Lemaire, P., et al. (2020).

MetNet: A Neural Weather Model for Precipitation

Forecasting.

Weyn, J. A., Durran, D. R., & Caruana, R. (2020).

Examined the ability of machine learning techniques to

be used to improve the current forecasting models of

the weather. Journal of Advances in Modeling Earth

Systems.

Weyn, J. A., Durran, D. R., & Caruana, R. (2021).

Improving data-driven global weather prediction using

deep convolutional neural networks on a cubed sphere.

Journal of Advances in Modeling Earth Systems.

Zhang, J., Wang, H., & Chen, X. (2021). Investigated the

application of CNNs in weather forecasting using

satellite images. IEEE Access, 9, 108759-108770.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

356