Content uploaded by Yagyanath Rimal
Author content
All content in this area was uploaded by Yagyanath Rimal on Jul 18, 2019
Content may be subject to copyright.
Vol. 03, No.02,(2019), 47-, ISSN:2581-3242
International Journal of Machine Learning and Networked
Collaborative Engineering
Journal Homepage: http://www.mlnce.net/home/index.html
DOI : https://doi.org/10.30991/IJMLNCE.2019v03i01.004
16
Machine Learning Prediction of Wikipedia Time Series
Data using: R Programming
Yagyanath Rimal
Faculty of Science and Technology
Pokhara University, Nepal
Email: rimal.yagya@pu.edu.np
Abstract
his review article explains the prediction of automatic learning of
Wikipedia time series data using r programming. Although many time
series forecast researchers have been analyzed the time series could not
cover the gap between chart interpretation and time series analysis of the
Internet database directly. Its main objective is to explain the simplest
way to time model series whose data structure was different using R
programming, the result was sufficiently summarized with different
forecast models. The simplest form of analysis with graphical interpretation to obtain conclusions from the
time search Cristiano_Ronaldo of Wikipedia, a best player in euro football team. Whose trend and prediction
is analyzed for next 2020 from the past records trend. Therefore, this document presents the simplest way to
predict time series data and its strengths for data analysis using R programming.
1. Introduction
A historical series is a set of observations at different measurement periods of an employee who collected at
regular time intervals, such as monthly, weekly or yearly, such as the state budget, etc. According to
(Gahirwal, 2018), the precondition is the one whose interval must be the same. Time series forecasts are
widely used in econometrics, mathematical and financial forecasts of statesmen and different weather
forecasts and earthquakes. Magazine without independent variable (VANNESCHI, 2017). With the time-
based model, the researcher can interpolate the graphics of the model and then foresee a future project. The
temporal variable usually uses yt = yt-1 + E as a univariate model. The data in cross section are such data
collection procedures that are similar to the time series, but the time series data only have one variable
depending on the interval and the association of a single value, but in the cross-data collection can collect
many variables Elements in a fixed time interval. The data model may vary based on the time series model.
It is not expected, although there are many types such as seasonal, trend, cyclic and random. Models are
increasing or decreasing models. The period of occurrence has been corrected within a year or less. The
cyclical model is a good example of the government budget. Some data are also random. Purely random,
whose average is zero and the variance is constant (CHEN, 2013). So the dispersion diagram will not
indicate the correct model. The automatic regressive model is a model that corresponds to the sequences yt-
1, yt-2 and t-3. Therefore, the model becomes yt = b0 + b1yt-1 + b2yt-2 + b3y3-3 ... ..Et. The AR (0) means
B0, so the AR model is better when it works. The moving average model always speaks of the error terms in
each regression and t = B0 + Et + Q1E1 + Q2Et-2 + QET-q. Therefore, ARMA is a precise model that uses
AR and MA models in temporal data (Michael Jachan, 2007). regressive automatic movement media
regressive integrated mobile media ARIMA is commonly known as the Box Jenkins methodology (1976), a
method used to predict the basis of information from its own variables, based on an analysis of trends
univariate tendencies. After analyzing the periodic properties of the variables, machine learning. Regulators,
T
Keywords
Data Analytics Machine
Learning, Autocorrelation
Function, Particle Automatic
Correlation Function
Ultimate Indoor Navigation: A Low Cost Indoor Positioning and Intelligent Path Finding
17
policy makers and companies to make serious and economic forecasts; however, the choice is based on the
two hypotheses (SALAM, 2013). The AR model is applied in series as stationary if there is an invariable
time variation. This decreases the major MA duration model applied to the autoregressive process
convergent final order that uses the autocorrelation function (ACF) is the covariance of the following terms
and functions automatic creation partial yt (PACF) and yt p. The series is not firm, can be fixed after
differentiation.
Machine learning is the system that takes input and output as input parameters, so that the computer system
automatically produces the above parameters based on some models. This can be applied to the analysis of
neural networks, deep learning and artificial intelligence networks. traditional programming always requires
two inputs and programs and possibly produces an output, while the process of machine learning always
requires an input and real requirement, since the formal entry system produces the show and expected output
depending on the needs of the system (Sunday 2018 ). A fixed series after an integrated differentiation in the
order 1. Analysis Box - Jenkins concerns a systematic method of identification, regulation, monitoring and
the use of integrated models of the integrated moving time series (ARIMA initial). The method is
appropriate for medium to long time series. (Buncher 2018) The best correspondence between the ARIMA
model and the time series data is the best correspondence of ARIMA (0,0,0) means that p = q = i = 0.
The data science process includes two processes; The first process begins with cleaning the raw data and
data collection analysis using different algorithms to produce the display data in this process. At each stage,
IT skills, mathematics, statistics and interpretation are required. However, the data available in today's world
in any structured, unstructured and semi-structured format are raw data. The primary phase includes the
integration of raw data and, therefore, the selection of the required data, at some point requires a processing
required before the analysis. data cleaning requires 50% to 80% of the work of a set of scientific data (Ruiz,
2017). Time series analysis includes methods for analyzing time series data to extract significant statistics
and other data characteristics. The prediction of time series is the use to predict future values based on the
values observed above. Historical headline series and retail sales in this publication are widely used for non-
stationary data such as the economic climate. We will demonstrate different approaches to predict the time
series to detail. There are two types of machine learning: supervised learning and unsupervised learning.
However, R programming, python and weka are the best tools for data analysis; The data scientist can use
many other data analysis processes. R has extensive facilities for analyzing time series data. This section
describes the creation of a historical series, seasonal decomposition, exponential models and modeling and
prediction with the ARIMA package the prognosis (Change, 2018). Modeling time series, as the name
suggests, means working with (time days, hours, minutes) time-based data for hidden information and
making informed decisions. Time series models are very useful models when data is related in series. Most
business houses that work with time series data to analyze the number of sales next year, website traffic,
place of competition and much more. However, it is also one of the areas that many analysts do not
understand. There are three basic criteria for a series to be classified as stationary series. The series average
should not be a function of time, but must be a constant. The variance of the series should not be a function
of time. This property is called homoscedasticity with a variable data distribution. The term covariance i-th
term and (i + m) th should not be a function of time, therefore, covariance is not constant over time for
uniform (Chohlan 2018). The reason I took this first section is that, unless the time series is stopped, it is not
possible to create a time series model. In cases where the fixed criterion is violated, the first requirement
becomes stationary time series and therefore to try out stochastic models to foresee this historical series.
There are many ways to bring this stationarity. This is the most basic concept of time series. (Srivastavo,
2015).
2. Using R Programing
Here I use data set of the database Cristiano_Ronaldo of Wikipedia, is the top scorer in international
football careers. He has won 26 trophies in his career, including five league titles, five UEFA Champions
League titles and the UEFA European Championship. Ronaldo, a highest scorer, holds the record for the
Master challenge.
International Journal of Machine Learning and Networked Collaborative Engineering, ISSN: 2581-3242
18
Source: Wikipedia, 2018
> install.packages("wikipediatrend")
Error in install.packages : Updating loaded packages
> library(wikipediatrend)
> devtools::install_github("petermeissner/wikipediatrend")
> library(wikipediatrend)
> wc=wp_trend( page="Cristiano_Ronaldo",from="2015-07-01",to="2018-10-31")
The package wp_trend() function that allows us to get page view of data time series data.
> wc
granularity date views
1124 daily 2018-0 ..36744 715 daily 2017-0 ..30132
612 daily 2017-0 ..19273 232 daily 2016-0 ..60506
927 daily 2018-0 ..23233 379 daily 2016-0 ..74485
1003 daily 2018-0 ..22562 454 daily 2016-0 ..18957
789 daily 2017-0 ..31208 994 daily 2018-0 ..38602 ... 1209 rows of data not shown
> library(ggplot2) > View(wc)
> qplot(date,views,data=wc)
Ultimate Indoor Navigation: A Low Cost Indoor Positioning and Intelligent Path Finding
19
From the above figure there were more seasonality in data sets but good in rising pattern in some dura
tion.
> summary(wc)
project language article
Length:1219 Length:1219 Length:1219 Class :character Class :character Class :character
Mode :character Mode :character Mode :character
access agent granularity
Length:1219 Length:1219 Length:1219
Class :character Class :character Class :character
Mode :character Mode :character Mode : character date views
Min.:2015-07-0100:00:Min.: 15478 1st Qu.:2016-04-30:01stQu.:23040
Median :2017-03-01: Median:27818
Mean:2017-03-01 00: Mean:37300
3rd Qu.:2017-12-30:3rd Qu.:37117
Max.:2018-10-31 00:Max.:519908
> wc$views[wc$views==0]=NA
> ds=wc$date
> y=log(wc$views)
> df=data.frame(ds,y)
> qplot(ds,y )
This figure plots the log goal score and data variable in x axis shows pattern.
> installed.packages("prophet")
> library(prophet)
Here we are using forecast function columns ds (date type) and y, the time series. If growth is logistic,
then df must also have a column cap that specifies the capacity at each ds. If not provided, then the model ob
International Journal of Machine Learning and Networked Collaborative Engineering, ISSN: 2581-3242
20
ject will be instantiated but not fit; use fit.prophet(m, df) to fit the model.
> m=prophet(df)
> m
$`growth`
[1] "linear"
$changepoints
[1] "2015-08-09 GMT" "2015-09-17 GMT" "2015-10-26 GMT"
[4] "2015-12-04 GMT" "2016-01-12 GMT" "2016-02-20 GMT"
………………………………
[22] "2017-11-04 GMT" "2017-12-13 GMT" "2018-01-21 GMT"
[25] "2018-03-01 GMT"
$n.changepoints [1] 25
$changepoint.range [1] 0.8
$yearly.seasonality [1] "auto"
$weekly.seasonality [1] "auto"
……………………….
$uncertainty.samples [1]1000
$specified.changepoints [1] FALSE
$start [1]"2015-07-01 GMT"
$y.scale [1] 13.16141
$logistic.floor[1] FALSE
$t.scale [1] 105235200
$changepoints.t
[1] 0.03201970 0.06403941 0.09605911 0.12807882 0.160
……………………..
[6] 0.19211823 0.22413793 0.25615764 0.28817734 0.320
[11] 0.35221675 0.38423645 0.41543514 0.44745484 0.47
[16] 0.51149425 0.54351396 0.57553366 0.60755337 0.63
[21] 0.67159278 0.70361248 0.73563218 0.76765189 0.79
$seasonalities
$seasonalities$`yearly`
$seasonalities$`yearly`[1]365.25
$seasonalities$`$fourier.[1] 10
Ultimate Indoor Navigation: A Low Cost Indoor Positioning and Intelligent Path Finding
21
$seasonalities$`yea$r.sca[1] 10
$seasonalities$`$mode[1]"addite"
$seasonalities$weekly$mode [1]
$extra_regressors list()
$`stan.fit` NULL
$params
$params$`k` [1] 0.6589777
$params$m [1] 0.7410152
$params$delta
[,1] [,2] [,3] [,4]
[1,] -6.962949e-05 -0.4095019 -0.3294997 -5.841874e-08
[,5] [,6] [,7] [,8]
[1,] -4.191081e-08 -3.546061e-09 2.497178e-07 0.0244269
………………
[,22] [,23] [,24] [,25]
[1,] -8.958495e-08 -3.950414e-08 -3.443782e-08 -0.0429
$params$sigma_obs [1] 0.02643767
$params$beta
[,1] [,2] [,3] [,4]
[1,] 0.005891249 -0.01776793 -0.008432092 0.01224181
[,5] [,6] [,7] [,8]
…………….
[,25] [,26]
[1,] 0.0001815386 0.001226727
$history
ds y floor t y_scaled
1 2015-07-01 10.129507 0 0.0000000000 0.7696371
2 2015-07-02 9.968854 0 0.0008210181 0.7574307
3 2015-07-03 9.834834 0 0.0016420361 0.7472479
………
198 2016-01-14 10.156656 0 01617405583 0.7716998
…………………………………………………………
199 2016-01-15 9.999298 0
0.1625615764 0.7597438
International Journal of Machine Learning and Networked Collaborative Engineering, ISSN: 2581-3242
22
200 2016-01-16 10.128190 0 0.1633825944 0.7695370
[ reached getOption("max.print") -- omitted 1019 rows ]
$history.dates
[1] "2015-07-01 GMT" "2015-07-02 GMT" "2015-07-03 GMT" [4] "2015-07-04 GMT" "2015-07-
05 GMT" "2015-07-06 GMT"
[7] "2015-07-07 GMT" "2015-07-08 GMT" "2015-07-09 GMT"………………………………………
…………….
[997] "2018-03-23 GMT" "2018-03-24 GMT" "2018-03-25 GMT"
[1000] "2018-03-26 GMT"
[ reached getOption("max.print") -- omitted 219 entries ]
[ reached getOption("max.print") -- omitted 219 entries ]
$train.component.cols
Additive weekly yearly multiplicati
1 1 0 1 0
2 1 0 1 0
………………………….
25 1 1 0 0
26 1 1 0 0
$component.modes
$component.modes$`additive`
[1] "yearly" "weekly"
[3] "additive_terms" "extra_regressors_additive"
[5] "holidays"
$component.modes$multiplicative
[1]"multiplicative_terms"
[2]"extra_regressorsmultiplicative"
attr(,"class")
[1] "prophet" "list"
> future=make_future_dataframe(m,periods=365) #For next year
> tail(future)
ds
1579 2019-10-26 1580 2019-10-27
1581 2019-10-28 1582 2019-10-29
1583 2019-10-30 1584 2019-10-31
Ultimate Indoor Navigation: A Low Cost Indoor Positioning and Intelligent Path Finding
23
> forecast=predict(m,future)
> tail(forecast)
ds trend additive additive_lowr
1579 2019-10-26 11.08094 -0.17044324 -0.17044324
1580 2019-10-27 11.08184 -0.08389419 -0.08389419
…………………..
additive_terms_weekly weeklylowr
1579-0.17044324 0.010662724 0.010662724
1580-0.08389419 0.093079248 0.093079248
……………………..
1584-0.18950107 -0.031694505 -0.031694505
weekly_upperyearly_lower upper
1579 0.010662724 -0.1811060 -0.1811060 -0.1811060
1580 0.093079248 -0.1769734 -0.1769734 -0.1769734
…………………….
1584 -0.031694505 -0.1578066 -0.1578066 -0.1578066
multiplicative_terms multiplicative
1579 0 0
………………
1584 0 0
multiplicative_terms yhat yhat
1579 0 10.11238 11.64221
………………..
1584 0 10.11592 11.63837
trend_lower trend_upper yhat
1579 10.50417 11.70324 10.91049
………………..
1584 10.49862 11.72232 10.89598
> tail(forecast[c('ds','yhat','yhat_lower','yhat_upper')])
ds yhat yhat_lower yhat_upper
1579 2019-26 10.91 10.11 11.642
…………………
1584 2019-31 10.89 10.11 11.638
> exp(10.89598)
International Journal of Machine Learning and Networked Collaborative Engineering, ISSN: 2581-3242
24
[1] 53959.01
> plot(m,forecast)
From the above plot Cristino Ronaldo had prosperous in upcoming years.
> prophet_plot_components(m,forecast)
From the above figure Cristiano Ronaldo had good future in real Madrid champions ship however Fri
day matches does not suite him and July is the best suitable moment.
Conclusion
A common goal of time series analysis is the extrapolation of past behavior in the future. Forecasting proced
ures include random walks, moving averages, trend models, simple, linear, quadratic, and seasonal exponenti
al time series models. Business forecasts can be based on historical data models used to predict future market
behavior. The time series forecasting method is a data analysis tool that measures historical data points, usin
g line charts to predict future conditions and events. It is essential to analyze trends before building any type
of time series model. The details that interest us refer to any kind of trend, seasonality or random behavior in
the series. Once we know that patterns, trends, cycles and seasonality, the function of the attacking forces an
d the defense of the competition could be analyzed, corrective measures will be taken, so Cristiano Ronaldo
will have a good future in the next years after entering in the Real Champions League of Madrid has will goo
d future coming day.
References
[1] Buncher, D. (2018). The Box-Jenkins methods. NCSS Statistical Software.
Ultimate Indoor Navigation: A Low Cost Indoor Positioning and Intelligent Path Finding
25
[2] Changing, S. L. (2018). The world, one article at a time. Toronto Canada. Opinion=my own.
ttps://www.linkedin.com/in/
susanli/, Sr. Data Scientist,.
[3] CHEN, M.-Y. (2013). Time Series Analysis (I). Department of Finance.
[4] Chohlan, A. (2018). A little book of R for time series . DataCamp's manipulating time series in R course
by Jeffrey Ryan. .
[5] Domingos, P. (2018). A Few Useful Things to Know about Machine Learning. Department of Computer
Science and Engineering.
[6] Gahirwal, M. (2018). Inter Time Series Sales Forecasting. Information Technology, Vivekanand
Education Information Technology, Vivekanand Education.
[7] Michael Jachan. (2007). Time-Frequency ARMA Models and Parameter Estimators for Underspread
Nonstationary. IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 55, NO. 9, SEPTEMBER 2007.
[8] Ruiz, A. (2017). The Cognitive Coder InfoWorld.
[9] SALAM, M. A. (2013). MODELING AND FORECASTING PAKISTNAN'S INFLATION BY USING.
Statistical Officer, Statistics Department, State Bank of Pakistan, Karachi, Pakistan.
[10] Srivastavo, T. (2015). A Complete Tutorial on Time Series Modeling in R.
[11] VANNESCHI, L. (2017). RETAIL FORECASTING UNDER THE INFLUENCE OF
PROMOTIONAL DISCOUNTS. Instituto Superior de Estatística e Gestão de Informação.
Author’s Biography
Yagyanath Rimal is lecturer of Computer Science Information Technology at
Pokhara University, Nepal. He received his Master of Science in Information
Technology from SMU in 2006, India with Information Technology and
programming specialization, he had wide area of interest like teaching, writing
books, doing research and publishing international research articles, web
development and presentation at international conferences.
How to Cite
Gunagwera, Alex, Kiani, Farzad, “Ultimate Indoor Navigation: A Low Cost Indoor Positioning and Intelligent Path
Finding”, International Journal of Machine Learning and Networked Collaborative Engineering, Vol. 03, No. 1, 2019, pp. 16-34.
doi : https://doi.org/10.30991/IJMLNCE.2019v03i01.002.
