Artificial Neural Network Based Water Qu
Artificial Neural Network Based Water Qu
Published By:
Retrieval Number F8841088619/2019©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijeat.F8841.088619 2778 & Sciences Publication
International Journal of Engineering and Advanced Technology (IJEAT)
ISSN: 2249 – 8958, Volume-8 Issue-6, August 2019
parameters [16], [17], [18], [19], [20], [21], [22], prediction B. Model Development: Named “River Water Quality
and forecasting with artificial neural networks: introduction, Forecasting Model” (RWQFM) using ANN in Weka
issues and applications [23], [24], time-series forecasting For development of the proposed model named here
using ANNs [25], [26], role of ANNs in ecological “RWQFM”, the authors have used the results of their
modelling [27], predicting stock price index [28], modelling research done in [35] where the best prediction accuracy of
electricity demand [29], forecasting of water quality using 95.9% was obtained when the value of the best learning rate
ANN [30], Modelling water quality [31], [32], time series is determined as 0.04 following 50000 epochs after a set of
forecasting of WQ of the river Godavari [33]. All such various experiments. As the learning rate (the amount the
research works leads to the conclusion that ANNs are weights are updated) and the number of epochs (the number
currently being used in finding the solution for prediction or of times the dataset presented to the network to train
forecasting problems in a variety of domains specifically in through) are the two main factors while performing training
the field of ecology. But a biggest research gap found by of the neural network. For each of the five stations authors
going through all these and other research literatures is that have taken the WQ dataset of size 180 for fifteen years
still there is no effort in the direction of modelling (2001-2015) from January to December. In order to decide
(predicting or forecasting) the water quality of the river the WQ classes the standards formulated by the Central
Ganga using the concept of ANNs as far as to the best Pollution Control Board (CPCB), India is followed where
knowledge of the authors of the present work. In summary, the quality of water is categorized into five different classes
the focus of this research is to develop a (RWQFM) river which is described as below:
water quality forecasting model for the river Ganga 1. Class ‘A’ : drinkable water
consisting four WQ parameters for every one of the five 2. Class ‘B’: bathing water only
testing sites along this river stretch by applying the 3. Class ‘C’: drinkable water but after traditional
technique of ANNs on the extensive time series remedy.
experimental dataset (monthly) for the period of fifteen 4. Class ‘D’: best only for marine and wild life
years from 2001 to 2015 using WEKA tool. 5. Class ‘E’ : only for agriculture and industries
III. RESEARCH METHODOLOGY For experimental purpose the initial file containing the input
data instances is prepared and stored in comma separated
A. Study Area and the Experimental Dataset Used value (csv) file. Then for preprocessing purpose in weka this
For the research purpose, the country wide holiest river of initial file is converted into another file format which is
India Ganga has been selected. The Ganga river is generated compatible with weka called attribute-relation file format
as a result of joining of River Bhagirathi with river (arff). A supervised machine learning approach called
Alaknanda at Devprayag starting up from Garhwal artificial neural network (ANN) is used for creating the
Himalayas.The Ganga River basin in India is highlighted in forecasting model by setting its parameters in such a way
Fig. 2. The monthly experimental data sets collection for the that its output performance will be efficient one. This can be
WQ of the River Ganga from 2001 to 2015 at five different achieved by conducting various experiments to develop the
stations named: Devprayag, Rishikesh, Haridwar, Jwalapur models and checking their performance using four different
and Roorkee in Uttarakhand analyzed by Limnology & metrics. The steps followed for development of ‘RWQFM’
Ecological Modelling Laboratory, Gurukul Kangri are as below:
Vishwavidhyalaya, Haridwar, Uttarakhand, India is being 1. Create csv files of water quality dataset of the
taken in this research. This data set comprised of four WQ Ganga River for previous fifteen years for all five
parameters temperature, pH, dissolved oxygen (DO) and stations.
biochemical oxygen demand (BOD). The goal of the current 2. Convert all csv files containing dataset to arff
research work is to develop a water quality forecasting format for suitable to run in Weka.
model for the river Ganga for each and every sampling 3. Import the data instances contained in arff file for
station along this river by applying the technique of ANNs particular station for preprocessing in Weka.
on the experimental dataset taken over the period of fifteen 4. Build the structure of ANN based time series
years from 2001 to 2015 using WEKA tool [34]. WEKA forecasting model in Weka using multilayer
tool provide support for different data mining problems perceptron as a classifier function by choosing best
involving classification, clustering and visualization. configuration of the network.
5. Run the forecasting process for the structured
forecasting model.
6. Save the run information and the evaluation results
for the model identified as best model by
comparing various forecaster outputs.
7. Save the developed forecasting model which comes
out to be best w.r.t. its performance metric.
The output view of experiments performed in Weka is as
shown in the Fig. 3.
Published By:
Retrieval Number F8841088619/2019©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijeat.F8841.088619 2779 & Sciences Publication
International Journal of Engineering and Advanced Technology (IJEAT)
ISSN: 2249 – 8958, Volume-8 Issue-6, August 2019
Sept-2015 177 1
Oct-2015 178 1
Nov-2015 179 1
Dec-2015 180 1
Jan-2016 181* 0.9967
Feb-2016 182* 0.9962
Mar-2016 183* 0.9983
Apr-2016 184* 1.0019
May-2016 185* 1.0052
June-2016 186* 1.009
July-2016 187* 1.0143
Fig. 3 Experimental View in Weka for output result Aug-2016 188* 1.0223
screen in developing the RWQFM Sept-2016 189* 1.0316
Oct-2016 190* 1.0405
IV. RESULTS AND ANALYSIS Nov-2016 191* 1.0486
For developing the proposed time series ‘RWQFM’, the Dec-2016 192* 1.0561
authors have developed the ANN based model to generate Thus, it is concluded from these results that at this site with
predictions (forecasts) for future monthly values (classes) of reference to the past fifteen years data the quality of Ganga
WQ at each and every station on the basis of known past water is extraordinary and is suitable for drinking purpose.
respective WQ values at corresponding station. Various The forecasting performance of developed model is
experiments were conducted and their results have been computed using various measures which is presented in the
analyzed by going through the steps mentioned above. We table II which showed that the developed model attained a
made monthly forecasts into the future for a one year period best accuracy of 99.52% as the minimum MSE is 0.0048
i.e. 12-steps ahead. The future forecasts/ predictions from during forecasting the WQ one month-ahead as highlighted.
end of training data is marked as ‘*’ that indicates the future
trends. Different metrics are used which perform relative Table-II: The forecasting performance of ANN model
measurement that compared the forecasted result to the last for station “Devprayag”
known target value i.e. class of WQ as the prediction which
is always taken as relative to the step at which the forecast is Performance MAE MAPE RMSE MSE
to be done. These matric are defined mathematically as Metric
follows:
1. Mean Absolute Error (MAE): sum(abs(predicted - Target WQ
Published By:
Retrieval Number F8841088619/2019©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijeat.F8841.088619 2780 & Sciences Publication
International Journal of Engineering and Advanced Technology (IJEAT)
ISSN: 2249 – 8958, Volume-8 Issue-6, August 2019
Table-III: 12 months-ahead forecasted result of ANN Fig. 5 Future Forecast for WQ (12-steps ahead) at
model (RWQFM) for Rishikesh station Rishikesh Station
Time Stamp Instance No. WQ Forecasting Model Development for Third Station
July-2015 175 1 “Haridwar”
Aug-2015 176 1 For the third station Haridwar the developed model resulted
Sept-2015 177 1 in the forecast for the WQ of the river Ganga for the next
Oct-2015 178 1 12-months in advance as given in the table V with ‘*’
Nov-2015 179 1 marked entries. The resulting forecast of WQ at this station
Dec-2015 180 1 is indicated graphically in the Fig. 6. It is concluded from
Jan-2016 181* 0.9938 these results that at this station the water is belonging to the
Feb-2016 182* 0.9975 class A everywhere except showing a little swing from this
Mar-2016 183* 0.9971 class towards class C and class B in the month of October
Apr-2016 184* 0.9941 and November. Therefore it is concluded that in general the
May-2016 185* 0.9941 water quality at Haridwar site is found suitable for drinking
June-2016 186* 1 but indicated a sign of little bit deterioration also particularly
July-2016 187* 0.9969 in two months. The developed model accomplished a best
Aug-2016 188* 0.9941 accuracy of 99.55% when forecasting the WQ in one step
Sept-2016 189* 1.0074 (month) advance as emphasized in table VI where the
Oct-2016 190* 0.9951 forecasting performance of developed model is presented.
Nov-2016 191* 0.9959
Dec-2016 192* 1.0113 Table-V: 12 months-ahead forecasted result of ANN
model (RWQFM) for Haridwar station
Table-IV: The forecasting performance of ANN model for Time Stamp Instance No. WQ
station “Rishikesh” July-2015 175 2
Aug-2015 176 2
Performance MAE MAPE RMSE MSE Sept-2015 177 2
Metric Oct-2015 178 1
Nov-2015 179 1
Target WQ Dec-2015 180 1
Published By:
Retrieval Number F8841088619/2019©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijeat.F8841.088619 2781 & Sciences Publication
International Journal of Engineering and Advanced Technology (IJEAT)
ISSN: 2249 – 8958, Volume-8 Issue-6, August 2019
Jan-2016 181* 1.9023 For the Jwalapur station the WQ of the river Ganga for the
Feb-2016 182* 0.6147 next 12 months (Jan 2016 to Dec 2016) in advance is
Mar-2016 183* 0.3812 presented in the table 7 with ‘*’ marked entries and in
Apr-2016 184* 1.582 finding this the forecasting performance of developed model
May-2016 185* 0.6254 is computed is mentioned in the table 8. The resulting
June-2016 186* 0.9188 forecast of WQ at this station is shown as a graph in the Fig.
July-2016 187* 1.0117 7 where a high fluctuation in the quality of water was
Aug-2016 188* 0.7937 noticed. Initially there exist an uncertainty in the water
Sept-2016 189* 0.103 quality and then in the last months the curve is showing that
the WQ is going to worsen to a greater extent.
Oct-2016 190* 3.5707
Consequently, authors concluded that at this place over-all
Nov-2016 191* 2.9232
the water is degrading momentously. The developed model
Dec-2016 192* -0.9287
achieved a best accuracy of 92.55% when it forecasts the
WQ one month-ahead as indicated in table 8.
Table-VI: The forecasting performance of ANN model for Table-VII: 12 months-ahead forecasted result of ANN
station “Haridwar” model (RWQFM) for Jwalapur station
Time Stamp Instance No. WQ
Performance MAE MAPE RMSE MSE
July-2015 175 4
Metric
Aug-2015 176 4
Sept-2015 177 4
Target WQ
Oct-2015 178 4
1 month- 0.0214 1.4842 0.0667 0.0045 Nov-2015 179 4
advance Dec-2015 180 4
2 month- 0.0541 3.3325 0.1707 0.0291 Jan-2016 181* -1.258
advance Feb-2016 182* 1.1846
3 month- 0.1093 7.4293 0.2855 0.0815 Mar-2016 183* -2.4063
advance Apr-2016 184* -1.794
4 month- 0.193 13.2483 0.5137 0.2639 May-2016 185* -3.4866
advance June-2016 186* -3.866
5 month- 0.3146 22.7623 0.7923 0.6277 July-2016 187* -2.6316
advance Aug-2016 188* -1.2381
6 month- 0.413 29.2842 0.943 0.8892 Sept-2016 189* 2.7471
advance Oct-2016 190* 5.8086
7 month- 0.5056 35.5322 1.057 1.1171 Nov-2016 191* 7.6206
advance Dec-2016 192* -2.3123
8 month- 0.6129 41.7054 1.1854 1.4052 Table-VIII: The forecasting performance of ANN model
advance for station “Jwalapur”
9 month- 0.8308 52.1879 1.5423 2.3785
advance Performance MAE MAPE RMSE MSE
10 month- 0.9121 53.4328 1.5797 2.4955 Metric
advance
11 month- 1.0178 56.7156 1.6452 2.7065 Target WQ
advance
12 month- 1.2289 68.9053 1.9621 3.8498 1 month- 0.1101 5.0165 0.2729 0.0745
advance advance
2 month- 0.2072 9.9251 0.4291 0.1841
advance
3 month- 0.3764 17.3772 0.7097 0.5037
advance
4 month- 0.6388 27.1995 1.1181 1.2501
advance
5 month- 1.0292 44.199 1.7836 3.1811
advance
6 month- 1.2967 53.4927 2.1733 4.7233
advance
7 month- 1.6712 66.6589 2.7312 7.4596
advance
Fig. 6 Future Forecast for WQ (12-steps ahead) at 8 month- 1.7452 76.9113 2.75 7.5626
Haridwar Station advance
Published By:
Retrieval Number F8841088619/2019©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijeat.F8841.088619 2782 & Sciences Publication
International Journal of Engineering and Advanced Technology (IJEAT)
ISSN: 2249 – 8958, Volume-8 Issue-6, August 2019
9 month- 1.9583 88.0516 2.9789 8.8736 1 month- 0.0395 3.033 0.0807 0.0065
advance advance
10 month- 2.2893 105.7638 3.2199 10.3677 2 month- 0.0484 3.4705 0.0905 0.0082
advance advance
11 month- 2.8557 137.3646 4.0213 16.1707 3 month- 0.0588 4.039 0.105 0.011
advance advance
12month- 2.8427 142.906 4.0343 16.2755 4 month- 0.0769 5.0519 0.1396 0.0195
advance advance
5 month- 0.1176 7.5158 0.244 0.0595
advance
6 month- 0.1574 9.267 0.3467 0.1202
advance
7 month- 0.2085 11.8011 0.4377 0.1916
advance
8 month- 0.3377 18.4448 0.7829 0.6129
advance
9 month- 0.3575 20.7527 0.7216 0.5206
advance
Fig. 7 Future Forecast for WQ (12-steps ahead) at 10 month- 0.4477 28.8266 0.8396 0.7049
Jwalapur Station advance
11 month- 0.5414 32.3212 0.9341 0.8725
Forecasting Model Development for Fifth Station advance
“Roorkee” 12 month- 0.6055 35.8225 1.0207 1.0419
For the Roorkee station the proposed model indicated advance
forecast for the WQ of the river ganga for the next 12-
months from the end of training data in the table IX with ‘*’
marked entries from January 2016 to December 2016 with
the forecasting curve shown in Fig. 8. It is concluded from
these results that the water quality at this station forecasted
to be of class A. The developed model given a best accuracy
of 99.35% when it forecasts the WQ one month-ahead as
highlighted in table X.
Published By:
Retrieval Number F8841088619/2019©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijeat.F8841.088619 2783 & Sciences Publication
International Journal of Engineering and Advanced Technology (IJEAT)
ISSN: 2249 – 8958, Volume-8 Issue-6, August 2019
the water at these sites is suitable for drinking purpose. Conference on Efficient & Sustainable Water Systems Management
toward Worth Living Development. Procedia Engineering Vol. 162,
However, at the Haridwar station in general, the water
pp. 153 – 161.
quality was also forecasted to be in class A i.e. suitable for 7. Krishna, B. et al. (2011). Time Series Modeling of River Flow Using
drinking but only in nine months while in the last months of Wavelet Neural Networks. Journal of Water Resource and Protection,
the year 2016 a minute degradation in quality was also Vol.3, pp. 50-59.
8. K. Abhishek, A. Kumar, R. Ranjan, S. Kumar, “A Rainfall
noticed. It can be because of the heavy tourist load.
Prediction Model using Artificial Neural Network”, Control and
Moreover, at the Jwalapur site the forecasting results System Graduate Research Colloquium (ICSGRC), IEEE, 2012.
indicated a lot of uncertainty as well as critical condition of 9. Imran Maqsood, Muhammad Riaz Khan, Ajith Abraham, “An
water quality in the last quarter of the year i.e. WQ is ensemble of neural networks for weather forecasting”, Neural Comput
degrading significantly. The reason may be heavy sewage & Applic, Springer-Verlag London Limited 4, 13: pp. 112–122, 2004.
10. Mohd. Fahmi Mohd. Nasir, H. Juahir, N.Roshan, I. Mohd. N.A.
dumping at this site. Shafie, N. Ramli, “Artificial Neural Networks Combined with
These results are on the basis of the forecasting performance Sensitivity Analysis as a Prediction Model for Water Quality Index in
performed by the developed model with accuracies: 99.52% Juru River, Malaysia”, International Journal of Environmental
Protection, IJEP Vol.1 No. 3 PP.1-8., 2011.
at Devprayag, 100% at Rishikesh, 99.55% at Haridwar,
11. N. M. Gazzaz, MohdKamilYusoff, Ahmad ZaharinAris b, Hafizan
92.55% at Jwalapur and 99.35% at Roorkee. At Rishikesh Juahir, Mohammad FiruzRamli, “Artificial neural network Modelling
station the WQ of river is astonishing or i.e. it will remain in of the water quality index for Kinta River (Malaysia) using water
class A (suitable for drinking). Among five stations at the quality variables as predictors”, Marine Pollution Bulletin,
ELSEVIER, 2012.
Rishikesh station the developed model attained a best
12. Sujana Prajith kumar, Shweta Verma, B V Mahajan, “Application of
forecasting accuracy of 100% having MSE=0 which means ANN model for the prediction of Water Quality Index ”, International
the developed model forecasts WQ effectively without any Journal of Engineering Research and General Science Volume 3,
error. Finally, the forecasted results show that the Issue 1, ISSN 2091-2730, January-February, 2015.
developed RWQFM (River Water Quality Forecasting 13. Faris Gorashi, Alias Abdullah, “Prediction of Water Quality Index
Using Back Propagation Network Algorithm. Case Study: Gombak
Model) based on ANN is more efficient in terms of the River”, Journal of Engineering Science and Technology Vol. 7, No. 4
accuracy. Consequently, the ANN based technique is proved (2012) 447 – 461, 2012.
to be highly proficient in the forecasting applications. Also, 14. Hossein Banejad, Ehsan Olyaie, “Application of an Artificial Neural
Network Model to Rivers Water Quality Indexes Prediction – A Case
in general the water of the Ganga river is found to be fine
Study”, Journal of American Science, 7(1) 2011.
among the undertaken stations in the year 2016. Thus, the 15. Sirilak Areerachakul, Siripun Sanguansintukul, “Water Quality
authors have got success in achieving their objective and Classification using Neural Networks: Case Study of Canals in
provided a support to the water management bodies for Bangkok, Thailand”, Institute of Electrical and Electronics Engineers,
handling the pollution issues related with the river water. 2009.
16. M. Heydari, E. Olyaie, H. Mohebzadeh and O. Kisi, “Development
Still this research work have few shortcomings like the of a Neural Network Technique for Prediction of Water Quality
available dataset size of fifteen years which can be Parameters in the Delaware River, Pennsylvania”,Middle-East Journal
extended, the study area and its geographical constraints of Scientific Research 13 (10): pp.1367-1376, 2013.
under consideration etc. As a future prospective firstly, the 17. Nibedita Guru, Ramakar Jha, “Simulation of BOD-DO Modelling in
Mahanadi River System lying in Odisha using ANN, India”, IOSR
same model can be enhanced by considering another Journal Of Environmental Science, Toxicology And Food
learning methods with extended data set and improved Technology (IOSR-JESTFT) ISSN: 2319-2402, ISBN: 2319-2399.
configuration of model parameters. Secondly, the Volume 2, Issue 4, PP 52-57, Jan. - Feb. 2013.
18. Vesna Rankovi, Jasna Radulovic, Ivana Radojevic,
forecasting can be done for other stations as well subject to
AleksandarOstojic, Ljiljana Comic, “Neural network Modelling of
the availability of historic quality data. In this way it is dissolved oxygen in the Gruza reservoir, Serbia”, Ecological
concluded that such models will be more proficient for Modelling 221 (2010) 1239–1244, ELSEVIR, 2010.
solving the forecasting problems in various application 19. Gholamreza Asadollahfrdi, Azadeh Hemati,Saber Moradinejad and
domains. Our results can be used by water management Rashin Asadollahfardi, “Sodium Adsorption Ratio (SAR) Prediction
of the Chalghazi River Using Artificial Neural Network (ANN) Iran”,
bodies which are concerned with pollution in rivers. Current World Environment Vol. 8(2), 169-178, 2013.
Therefore, this research work can really be fruitful for the 20. M. Niroobakhsh, S. H. Musavi-Jahromi, M. Manshouri, H. Sedghi,
society and for the entire nation as well. “Prediction of water quality parameter in Jajrood River basin:
Application of multi-layer perceptron (MLP) perceptron and radial
basis function networks of artificial neural networks (ANNs) ”,
REFERENCES African Journal of Agricultural Research Vol. 7(29), pp. 4131-4139,
31 July, 2012.
1. R. Dey, V. Bajpai, G. Gandhi and B. Dey, “Application of Artificial 21. Archana Sarkar, Prashant Pandey, “River Water Quality Modelling
Neural Network (ANN) technique for Diagnosing Diabetes Mellitus”, using Artificial Neural Network Technique”, International Conference
IEEE Kharagpur Section &IEEE Sri Lanka Section Region 10 on Water Resources And Ocean Engineering (ICWRCOE),
Colloquium and the Third ICIIS, Kharagpur, INDIA December 8-10, ELSEVIER, 2015.
2008. 22. Anita Talib, Mawar Idati Amat, “Prediction of Chemical Oxygen
2. S. Chattopadhyay, “A neuro-fuzzy approach for the diagnosis of Demand In Dondang River Using Artificial Neural Network”,
depression”, Applied Computing and Informatics, article in press, International Journal of Information and Education Technology, Vol.
Production and hosting by ELSEVIER, 2014. 2, No. 3, June 2012.
3. C. Mahesh, V. G. Suresh, Manjula Babu, “Diagnosing Hepatitis B 23. G. Zhang, B. Eddy Patuwo, Michael Y. Hu, “Forecasting with
Using Artificial Neural Network Based Expert System”, International artificial neural networks: The state of the art”, International
Journal of Engineering and Innovative Technology (IJEIT) Volume 3, Journal of Forecasting 14, ELSEVIER, pp. 35–62, 1998.
Issue 6, December 2013. 24. H. R. Maier, Graeme C. Dandy, “Neural networks for the prediction
4. S. Riad & J. Mania (2004). Rainfall-Runoff Model Using an and forecasting of water resources variables: a review of modeling
Artificial Neural Network Approach. Mathematical and issues and applications”, Environmental Modeling & Software 15,
Computer Modelling 40 (2004) pp. 839-846. ELSEVIER, pp. 101–124, 2000.
5. Abrahart, R.J. & See, L.M (2007). Neural network emulation of a
rainfall-runoff model. The journal of Hydrology and Earth System
Sciences 4, 287–326.
6. Reyes J.V. et al. (2016). Artificial Neural Networks applied to flow
prediction: A use case for the Tomebamba river. In International
Published By:
Retrieval Number F8841088619/2019©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijeat.F8841.088619 2784 & Sciences Publication
International Journal of Engineering and Advanced Technology (IJEAT)
ISSN: 2249 – 8958, Volume-8 Issue-6, August 2019
AUTHORS PROFILE
Published By:
Retrieval Number F8841088619/2019©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijeat.F8841.088619 2785 & Sciences Publication