High-Level Project Summary
Prediction of weather condition is important to take efficient decisions. In general, the relationship between the input weather parameters and the output weather condition is non linear and predicting the weather conditions in non linear relationship posses challenging task. The traditional methods of weather prediction sometimes deviate in predicting the weather conditions due non linear relationship between the input features and output condition. Motivated with this factor, we propose a neural networks based model for weather prediction. The superiority of the proposed model tested with the weather data collected from Indian metrological Department . weather prediction
Link to Final Project
Link to Project "Demo"
Detailed Project Description
INTRODUCTION
Weather forecasting is the use of science and technology to
predict the atmospheric conditions of a given area and time.
People have been trying to predict the weather informally for
thousands of years and since the 19th century. Weather
forecasts are made by collecting information about the current
state of the atmosphere in a particular area and then using the
weather to predict how the atmosphere will change. Individual
input is still required to select the best predictive model to
establish the prediction
When it comes to human activity that is largely based on
changes in barometric stress, current climate, and weather or
cloud cover, weather forecasting is now relying on computer-
based models that look at a number of celestial objects.
Individual input is still required to select the best predictive
model to establish the prediction, which includes pattern
recognition skills, telephone communication, model
performance information, and model bias information.
The use of computers in the field of information management
is well known to us .The use of computers in the university
management system offers the following benefits in addition
to the manual system.
Space Agency Data
Dataset from Weather Station
We collected data on the actual weather of Nashville city from
wunderground.com, as well as nine other cities around
Nashville: Knoxville, Chattanooga, Jackson, Bowling Green,
Paducah, Birmingham, Atlanta, Florence, and Tupelo. By
finding location and date, the wunderground API returns a list
of weather view data
Machine Learning Techniques
In this study, since the predicted results are continuous
numerical values, the temperature in us, we use the regression
method. We find that Random Forest Regression (RFR) is a
superior regressor, as it incorporates many decision trees
while making a decision. In addition, we suggest comparing
several other state-of-the-art ML strategies with the RFR
process. Restoration strategies included Ridge Regression
(Ridge), Support Vector (SVR), Multi-layer Perceptron
(MLPR), and Extra-Tree Regression (ETR).
Dataset from Weather Station
We collected data on the actual weather of Nashville city from
wunderground.com, as well as nine other cities around
Nashville: Knoxville, Chattanooga, Jackson, Bowling Green,
Paducah, Birmingham, Atlanta, Florence, and Tupelo. By
finding location and date, the wunderground API returns a list
of weather view data.
(a) RMSE on test set while considering neighboring
cities
(b) RMSE on test set with increasing training size
(c) RMSE on test set for different ML models
Data Preprocessing
After receiving raw data from 'underground', we make sure that
each line (record) in the database contains records of all ten
cities for a period of time. We eliminate any feature with
empty or invalid data while creating a database. Also, we
convert separate elements in the database, such as wind
direction and position, into dummy / indicator variables using a
process called 'One Hot Encoding'.
We underwent this modification prior to the division of
training and evaluation data. This is because, in both training
and testing data, we need the same amount of feature
variability. If we make this adjustment after a split, then there
is no guarantee that both will have all the values of the
categories of adjustable features.
If the number of training phase values and test sets are not the
same, the conversion reflects the number of different features
of these sets. That’s why we need to make this transition before
the division of training and database testing.
In addition, we do measure
x
←
x
σ
−
µ
mean to all used as a
training set, that week is the previous week of continuous
variables so that the variables have almost zero meaning,
which works, reducing computer costs while training models.
Hackathon Journey
Choose hackathons where you have an idea and you can help in providing a solution. If you are still unclear about the idea , then you can refer to previous implementations of that problem on internet to see how people have tackled it. Do not copy ideas directly from internet. Hackathons are about your own ideas and how you implement it
References
[1] Y. Xu, V. Ramanathan, D. G. Victor, Global warming
will happen faster than we think, Nature, Vol. 564,
30:32, 2018, Dec.
[2] D. Spratt, I. Dunlop, Existential climate-related
security risk: A scenario approach, 2019 May, Policy
Paper from Breakthrough-National Centre for Climate
Restoration
[3] Aditya Grover, Ashish Kapoor, and Eric Horvitz.
2015. A deep hybrid model for weather forecasting.
In Proceedings of the 21th ACM SIGKDD
International Conference on Knowledge Discovery
and Data Mining. ACM, 379–386.