WEATHER PREDICTION FOR INDIAN LOCATION USING MACHINE LEARNING

Transcription

1 Volume 118 No , ISSN: (on-line version) url: ijpam.eu WEATHER PREDICTION FOR INDIAN LOCATION USING MACHINE LEARNING 1 Jitcha Shivang, 2 S.S Sridhar 1 Student, Department of Computer Science and Engineering SRM IST 2 Professor, Department of Computer Science and Engineering SRM IST 1 shivang_jitul@srmuniv.edu.in, 2 sridhar.s@ktr.srmuniv.ac.in Abstract: In this paper, a simulated system is developed to predict various weather conditions across Indian subcontinent using Data Analysis and Machine learning techniques such as linear regression and logistic regression. The main source of data to be used for supervised learning is to be collected from data.gov.in, ncdc.noaa.gov and UCI machine learning data repository. The existing weather condition parameters ex. temperature etc are used to fit a model and further using machine learning techniques and extrapolating the information, the future variations in the parameters are analysed. Keywords: weather, climate, forecast, python, prediction, linear regression, machine learning. 1. Introduction Traditionally, weather forecasting has always been performed by physically simulating the atmosphere as a fluid. The current state of the atmosphere is sampled. The future state of the atmosphere is computed by solving numerical equations of thermodynamics and fluid dynamics. But this traditional system of di erential equations that govern the physical model is sometimes unstable under disturbances and uncertainities while measuring the initial conditions of the atmosphere. This leads to an incomplete understanding of the atmospheric processes, so it restricts weather prediction up to a 10 day period, because beyond that weather forecasts are significantly unreliable.but Machine learning is relatively robust to most atmospheric disturbances as compared to traditional methods. Another advantage of machine learning is that it is not dependent on the physical laws of atmospheric processes. Machine Learning Techniques temperatures. Linear regression is not used for weather classification of each day because this algorithm cannot be used with classification data. Functional Regression: The second algorithm to be used is a type of functional regression. It looks for historical weather patterns which are similar to the present day weatherpatterns, then it predicts the future weather condition based upon the data of the historical weather patterns. 2. Literature Survey Mark Holmstrom, Dylan Liu, Christopher Vo (2016) concluded that both linear and functional regression did not perform as well as professional weather forecasting methods but in the longer run differences in their performances decreased, suggesting that over a longer period of time, Machine learning can indeed outperform professional and traditional methods.linear regression is a low bias and high variance algorithm and henceits accuracy can be improved by collecting further data. PiyushKapoor and Sarabjeet Singh Bedi (2013) concluded that if we performcomparison of weather condition variation by sliding window algorithm, the results are highly accurate except for the months of seasonal change. The results can be altered by changing the size of the window. Accuracy of the unpredictable months can be increased by increasing the window size to one month. DivyaChauhan and Jawahar Thakur (2013) made a comparison in their paper, which shows that the algorithms such as k-mean clustering and decision trees are well suited for mining data to predict future weather conditions. If we increase the size of the training set, the accuracy at first increases but then it slowly decreases after a particular period of time, depending on the size of the dataset. Linear Regression: it uses all the features present in the dataset and gives a linear graph combining high and low 1945

2 Qing Yi Feng1,RuggeroVasile, Marc Segond, AviGozolchiani, Yang Wang, Markus Abel, ShilomoHavlin, Armin Bunde, and Henk A. Dijkstra1(2016) have made a machine-learning toolbox which is based on climate data gathered from analysis and reconstruction of complex networks. It can also handle data containing multiple variables from these networks. The development of predictor models in the toolbox is dynamic and data-driven. Siddharth S. Bhatkande, Roopa G. Hubballi(2016) In their work the authors have used data mining technique and Decision tree algorithm as a means to classify weather parameters like maximum temperature, minimum temperature in terms of day, month and year. Sanyam Gupta, Indumathy, GovindSinghal (2016)suggested and proposed an efficient and accurate weather prediction and forecasting model using linear regression concepts and normal equation model. All these concepts are a part of machine learning. The normal equation is a very efficient weather prediction model and using the entities temperature, humidity and dew-point, it can be used to make reliable weather predictions. This model also facilitates decision making in day to day life. It can yield better results when applied to cleaner and larger datasets. Muthulakshmi A, ME (SE), Dr.S.Baghavathi Priya(2015) in their work proposed a methodology that aims at providing an efficient and accurate weather forecasting models to predict and monitor the weather datasets to predict rainfall. In the past, the parameters of weather were recorded only for the present time. But in the future, work will be done to make a working model of selection that can be used for classifying the framework for continuous monitoring of the climatic attributes. Aditya Grover, AshishKapoor and Eric Horvitz in their work made a weather prediction model that predicts by considering the joint influence of key weather variables. They also made a kernel and showed that interpolation of space can be done by using GPS with such a kernel, taking into account various weather phenomena like turbulence. They also performed temporal analysis within a learner based on gradient tree and augmented the system using deep neural network. John K. Williams and D. A. Ahijevych, C. J. Kessinger, T. R. Saxen, M. Steiner and S. Dettlinghave shown in their work that a set of skilful predictors for thunderstorm initiation can be identified by using the random forest machine learning algorithm. The random forest method can also be used to identify regimes in which they can improve the skill of the application by using a forecast logic. 3. Proposed Model The proposed model will use linear regression, which will predict the high and low temperatures as a linear combination of all the features. Linear regression does not use weather classification data of each day because this algorithm cannot be used with classification data. Therefore initially in our project only eight parameters are selected for use which aremaximum temperature, minimum temperature, mean humidity, and mean atmospheric The second algorithm to be used is a type of functional regression. It looks for historical weather patterns which are similar to the present day weatherpatterns, then it predicts the future weather condition based upon the data of the historical weather patterns. The architecture of the proposed model is given in Fig.1. A Neural network model is used to train 80% of data values and remaining 20% for testing. Algorithm 1. Train model with x (i) where x (i) W. W={ mxt, mnt, meanh, meanap} of past few years 2. Evaluate and optimize the model with test set 3. a. 4-fold cross validate the model with blind set b. Minimize the cost function 4. Input past two days mxt and mnt (after the model is ready) 5. Predict the mxt and mnt for the next day *mxt=maximum temperature, mnt=minimum temperature, meanh=mean humidity, meanap= mean atmospheric pressure. 1946

3 4. Architecture of Proposed Model Figure 1. Architecture of Proposed Model 5. Result 6. Conclusion In our work, both linear regression and functional regression are used to predict the weather parameters. The same dataset is used in both the algorithms, so that a comparative analysis could be made. Forecasting weather parameters for a longer duration with more parameters involves the use of artificial Neural Network. To check the validation of prediction model for weather conditions, both systems are compared to check the fitness of applicability. The future work involves building a prediction model using Deep Neural Network. References [1] Mark Holmstrom, Dylan Liu, Christopher Vo, Machine Learning Applied to Weather Forecasting, Stanford University, [2] PiyushKapoor and Sarabjeet Singh Bedi Weather Forecasting Using Sliding Window Algorithm, Kvantum Inc., Gurgaon , India MJP Rohilkhand University, Bareilly , India, [3] DivyaChauhan, Jawahar Thakur Data Mining Techniques for Weather Prediction: A Review, Shimla 5, India: ISSN,

4 [4] Qing Yi Feng1, RuggeroVasile,Marc Segond4, AviGozolchiani, Yang Wang, Markus Abel, ShilomoHavlin, Armin Bunde, and Henk A. Dijkstra1 ClimateLearn: A machine-learning approach for climate prediction using network measures,, Germany, [5] Siddharth S. Bhatkande1,Roopa G. Hubballi2 Weather Prediction Based on Decision Tree Algorithm Using Data Mining Techniques, Belgaum India: International Journal of Advanced Research in Computer and Communication Engineering, [6] Sanyam Gupta, Indumathy, GovindSinghal Weather Prediction Using Normal Equation Method and Linear regression Techniques, Vellore, Tamil Nadu, India: International Journal of Computer Science and Information Technologies, [7] Muthulakshmi A, ME (SE), Dr.S.BaghavathiPriya A survey on weather forecasting to predict rainfall using big data analytics, Chennai.: IJISET, [8] Aditya Grover, AshishKapoor, Eric Horvitz (n.d.) A Deep Hybrid Model for Weather Forecasting, : Microsoft Research, Redmond. [9] John K. Williams and D. A. Ahijevych, C. J. Kessinger, T. R. Saxen, M. Steiner and S. Dettling National Center for Atmospheric Research, Boulder, Colorado (n.d.) A MACHINE LEARNING APPROACH TO FINDING WEATHER REGIMES AND SKILLFUL PREDICTOR COMBINATIONS FOR SHORT-TERM STORM FORECASTING, Colorado. [10] S.V.Manikanthan and T.Padmapriya Recent Trends In M2m Communications In 4g Networks And Evolution Towards 5g, International Journal of Pure and Applied Mathematics, ISSN NO: , Vol-115, Issue -8, Sep [11] S.V. Manikanthan, T.Padmapriya An enhanced distributed evolved node-b architecture in 5G telecommunications network International Journal of Engineering & Technology (UAE), Vol 7 Issues No (2.8) (2018) March [12] S.V. Manikanthan, T. Padmapriya, Relay Based Architecture For Energy Perceptive For Mobile Adhoc Networks, Advances and Applications in Mathematical Sciences, Volume 17, Issue 1, November 2017, Pages

5 1949

6 1950