Override Traditional Decision Support Systems
How Trajectory ELT Processes Modeling Improves Decision Making?
Noura Azaiez
1
and Jalel Akaichi
2
1
Department of Computer Science, Institut Supérieur de Gestion, University of Tunis, Tunis, Tunisia
2
Department of Computer Science, College of Computer Science, King Khalid University, Abha, Saudi Arabia
Keywords: Trajectory ELT Processes, Extraction, Loading, Transformation, Trajectory Construction, Trajectory Data
Source Model, Trajectory Data Mart Model, Model Driven Architecture.
Abstract: Business Intelligence is often described as a set of techniques serving the transformation of raw data into
meaningful information for business analysis purposes. Thanks to the technology development in the realm
of Geographical Information Systems, the so-called trajectory data were appeared. Analysing these raw
trajectory data coming from the movements of mobile objects requires their transformation into decisional
data. Usually, the Extraction-Transformation-Loading (ETL) process ensures this task. However, it seems
inadequate to support trajectory data. Integrating the trajectory aspects gives the birth of Trajectory ETL
process (T-ETL). Unfortunately, this is not enough. In fact, the business analysis main purpose is to
minimize costs and time consuming. Thus, we propose to swap the T-ETL tasks scheduling: instead of
transforming the data before they are written, the Trajectory Extraction, Loading and Transformation (T-
ELT) process leverages the target system to achieve the transformation task. In this paper, we rely on a set
of powerful mechanisms to handle the complexity of each T-ELT task. Wherefore, an algorithm is dedicated
to ensure the transformation of raw mobile object positions into trajectories and from there we highlight the
power of the Model-driven Architecture approach to transform the resulting trajectories into analytical data
in order to perform the Business Intelligence goal.
1 INTRODUCTION
Often Business Intelligence (BI) applications use
data gathered from data warehouse (DW) to support
a wide range of business decisions ranging from
operational to strategic. Data warehousing concept
evolves according to technological developments. In
fact, the incredible progress related to Geographical
Information Systems and positioning technologies
imposes moving beyond the traditional to provide
the real-time tracking of the movement of mobile
objects and generate a new data kind called
Trajectory Data (TD). TD collected from trajectory
sources reflect the moves and stops of the mobile
objects in the real world. However, the trajectory
operational data sources are poorly suited to long-
term vision and therefore seem inadequate for
decision making. Towards this inadequacy, the
notion of Trajectory Data Warehouses (TDWs) was
appeared (Azaiez and Akaichi, 2016) to support TD.
Before their availability for decisional purposes, raw
TD have to flow through a set of tasks under the
general title Extract-Transform-Load trajectory ETL
process. This latter is responsible for extracting data
from heterogeneous sources, cleaning them
according to business rules, and loading the
standardized data into a TDW. However, it is not
enough. In fact, the business analysis main purpose
is to minimize costs and time consuming. For that,
we propose to swap the ETL tasks scheduling by
isolating the extract and load processes from the
transformation process. Instead of transforming the
data before they are written, Trajectory Extract,
Load and Transform (T-ELT) processes leverages
the target system to perform the transformation task.
In this paper, we rely on a set of powerful
mechanisms that ensure the better data
transformation, starting from extracting TD from
sources, loading them to reach the transformation
task that has to occur in the target area. Since the
transformation task is the core of the T-ELT
processes, we divide the running of this task into
two stages. The first sub-task offers an algorithm
that focuses on transforming trajectory raw
550
Azaiez N. and Akaichi J.
Override Traditional Decision Support Systems - How Trajectory ELT Processes Modeling Improves Decision Making?.
DOI: 10.5220/0006269605500555
In Proceedings of the 5th International Conference on Model-Driven Engineering and Software Development (MODELSWARD 2017), pages 550-555
ISBN: 978-989-758-210-3
Copyright
c
2017 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
positions, describing mobile objects movements,
into full trajectories. The generated trajectories are
stored in a Trajectory Database (TDB) as a
Trajectory Data Source model (TDSrc). For
decisional purposes, we emphasize on the power of
the Model Driven Architecture (MDA) (OMG,
2004) to transform TDSrc into Trajectory Data Mart
(TDM) models. This presents the second
transformation sub-task goal.
To reach goals scheduled above, we propose to
organize this paper as follow. In section 2, we
discuss some research works that treat the ETL and
T-ETL process modeling. Section 3 presents our T-
ELT proposed framework. Section 4 is dedicated to
illustrate our approach by an Epilepsy patient states
case study. We conclude the paper in section 5.
2 STATE OF THE ART
ETL processes emerged to provide a development
environment for maintenance, analyzing and
discovering. In this section, we present some
practical solutions proposed by the research
community that ETL design presents its area of
interest.
2.1 ETL Processes Modeling
Data Warehousing emerged to collect and structure
data at the aim to be the process of a good decision
making. ETL processes present a key component of
Data Warehousing since misleading data may
produce wrong business decisions. Despite its
importance, few are the research works that dealt
with the modeling of ETL processes.
The first attempt towards a conceptual model
dedicated to the design of the data warehouse
refreshment was presented in (Vassiliadis et al.,
2003). To deal with the ETL modeling problem,
authors proposed a generic meta-model for the
definition of the data centric part of ETL activities.
The customization and the extensibility of the meta-
model building allows designer to enrich it with his
own re-occurring patterns. Basing on their proposed
meta-model, a reusable framework named ARKTOS
II is introduced. ARKTOS II is complemented in
(Simitsis, 2003) wherein the author investigated the
ETL modeling problem relying on a method that
treats the collection of requirements and analyzes the
structure and the content of the existing data sources
and, their intentional mapping to the common DW.
Another manner to deal with the ETL modeling
is offered in (Boussaid et al., 2003) wherein authors
proposed a new approach for complex data
integration, based on a Multi-Agent System (MAS).
MAS is composed of a set of intelligent agents which
are responsible to achieve the major tasks of ETL
modeling.
In order to create an active ETL tool, Rifaieh and
Benharkat (2002) specified a model based on queries
to represent mapping guideline and expressions
between the source and the target data. They proved
this model by creating a warehousing tool (QELT).
2.2 Trajectory ETL Processes
Modeling
As it is presented above, several approaches were
proposed to solve the problem of the ETL processes
modeling. Those solutions seem unable to support
Trajectory Data. The research work (Zekri and
Akaichi, 2014) dealt with this problem and
presented a Trajectory ETL model that is able to
clean and manage Trajectory Data. Authors relied on
the UML language in order to describe the ETL
processes as a flow of activities. Authors focused on
the TDW conceptual modeling in order to facilitate
the analysis of Trajectory Data in the
multidimensional context. In their case, they need to
analyze the activities of a mobile medical delegate.
Hence, they proposed two algorithms to implement
trajectory ETL tasks and construct trajectories.
Marketos et al. (2008) investigated the problem
of ETL modeling for TDW design. Their work
supports a set of steps for building a real world
TDW, from trajectory reconstruction to data cube
feeding and aggregating over summary information.
They relied on a set of algorithms and equations in
order to extract measures that are able to handle the
ETL processes modelling.
2.3 Discussion
The literature meets several research works that
investigated the ETL modeling in different manners.
Following the study of the sample described above,
it becomes possible to discuss the strengths and
weaknesses of each one. According to a depth study
in ETL design field, three major approaches are the
basis of authors works: ETL modeling based on a)
conceptual constructs (Vassiliadis et al., 2003;
Simitsis, 2003), b) UML environment (Boussaid et
al., 2003; Zekri and Akaichi, 2014) and, c) mapping
expressions and guidelines (Rifaieh and Benharkat,
2002). The ETL processes scheduling is the
common point that the majority of works relied on.
In fact, authors investigated the problem of the ETL
Override Traditional Decision Support Systems - How Trajectory ELT Processes Modeling Improves Decision Making?
551
modeling basing on the same strategy; Extraction,
Transformation and then Loading. Relying on the
ETL processes is a gainful strategy at the time
consuming level. In fact, only data relevant to the
solution are extracted and processed. Hence, the
target area contains only data relevant to the
presentation. This potentially reduce development
and therefore time consuming; reduced warehouse
content simplifies the security regime implemented
and thus the administration overhead. However, the
ETL strategy complains from several weaknesses
especially the lack of flexibility which is explained
by the fact of targeting only relevant data for output.
That means that any future requirements, which may
need data that weren’t included in the original
design, must be added to the ETL routines. As a
result, this increases time and costs involved.
Integrating the concept of mobility into the
professional realm offers the possibility of reducing
geographical disparities related to organization
services. However, few are the works that dealt with
the Trajectory ETL problem as (Zekri and Akaichi,
2014) wherein authors relied on the UML language
in order to describe the ETL scenario, and (Marketos
et al., 2008) in which authors relied on a measures
to generate algorithms that serve the trajectory ETL
handling. The proposed frameworks serving the
Trajectory ETL process modeling are complex and
need high level experts to handle all tasks. Thus, we
propose a new approach that aims at moving beyond
Traditional DSS in order to enhance extracted
knowledge quality. After taking a look on the works
relying on the MDA approach, we note that it
expresses its powerful features in some areas such as
the DW evolution (Taktak et al., 2015) wherein
authors relied on the MDA approach to automate the
propagation of the evolutions occurred in the source
database towards the multidimensional DW.
3 OUR APPROACH OVERVIEW
Thanks to the Trajectory ELT process strategy, all
data, reflecting the moves and stops of mobile
objects in the real world, are extracted and loaded
into the warehouse. This, combined with the
isolation of the transformation process, offers an
easy feasibility of incorporating future requirements
into the DW structure; what proves the future
efficiency of the ELT process. Minimizing risks is
another issue that can be solved relying on ELT
process. In fact, isolating the load process from the
transformation process removes an inherent
dependency between each stage of the DW build
processes. This provides an excellent platform for
maintenance. As a result, basing on ELT process
presents a tremendous payoff strategy for decision
making systems. Our proposed framework relies on
the Trajectory ELT (T-ELT) process (Figure 1) to
emphasize a new gainful strategy that aims at
transforming trajectory raw data into trajectory
decisional data.
Figure 1: Our approach framework (T-ELT).
As the T-ELT is considered as a flow of
activities, we choose to begin this flow with the
identification of geographic points. To do so, we
focus on extracting (E) data from external sources.
The extraction method chosen by decision makers is
highly dependent on the source system. While
speaking about trajectory data, we have to extract
trajectory data that describe mobile object activities
from external trajectory sources. A generator tool is
responsible to perform this task. Note that collecting
these data requires its equipping with a positioning
system as GPS… We call trajectory raw data the
data as captured from the positioning device.
However, these data are not available for analysis.
Here, the main role of the Loading task (L) emerges.
In effect, the loading task is responsible to load all
extracted TD in the target server to be ready for the
required transformations. Achieving this task cannot
be ensured by traditional loading tools. Thus, ELT
works with high-end data engines such as Hadoop
clusters. In fact, Hadoop Distributed File System
(HDFS) is able to store everything from structured
and unstructured data. Relying on Hadoop
technology was not randomly chosen. It presents a
profitable strategy from all sides. Starting from
rapidity that is required in our work. Hadoop clusters
are known for boosting the speed of data analysis
applications. If a cluster's processing power is
overwhelmed by growing volumes of data,
additional cluster nodes can be added to increase
throughput. Besides, Hadoop clusters also are highly
resistant to failure because each piece of data is
copied onto other cluster nodes; which ensures that
data are not lost if one node fails. As Hadoop
features are conform to the loading goal, espacially
MODELSWARD 2017 - 5th International Conference on Model-Driven Engineering and Software Development
552
for ELT process, we rely on this technology to load
all gathered raw TD in the warehouse area.
For decisional purposes, TD recently loaded
have to be transformed. Due to its complexity, we
propose to divide the transformation process into
two sub tasks: T1 and T2. T1 purpose is to convert
the unstructured positions received by positioning
devices into structured TD in order to describe the
continuous objects movements. So, we propose a
trajectory construction algorithm that transforms
sequences of raw time-stamped positions location
data into meaningful trajectories that are then stored
into a Trajectory Database (TDB) as a Trajectory
Data Source model (TDSrc).
Trajectory construction algorithm description
Raw points arrive in bulk sets. Each point P (x,
y, t) has to be affected to its appropriate Mobile
Object (MO). Hence, each time the algorithm meets
a new position (P
new
), it checks whether the MO has
been processed so far. If so:
We have to check if this new point (P
new
)
satisfy a set of conditions to be added to the existing
partial trajectory (T) of MO or to create a new
trajectory (T
new
) starting point. For that, we inspire
from the work of (Marketos et al., 2008) and (Zekri
and Akaichi, 2014) to use the two generic
parameters “gap
time
” and “gap
space
”. gap
time
presents
the maximum time allowed to connect the existing
trajectory last point (L) to P
new
. The same for gap
space
that pressents the maximum distance allowed to
connect L to P
new
. Exceeding the gap
time
and gap
space
creates a new trajectory T
new
, and P
new
presents the
starting point of T
new
. Here, we spoke about the
trajectories organization. This step is ensured by the
function check_point_state (P
new
, L, gap
space
,
gap
time
) which verify the distance between P
new
and
the last point L of the existing trajectory T. If this
distance is lower than the gap
time
and gap
space
then
P
new
will be connected to L. Otherwise, P
new
becomes
the starting point of a new trajectory T
new
.
Once P
new
satisfies the conditions
presented above and it is added to the existing
trajectory T that describes a mobile object MO
activities, the function check_linearity (L.LastPoint,
L, P
new
) checks if P
new
forms a straight segment with
the two last points that are just before P
new
. If so, the
point in the middle L will be removed since the MO
has already passed through this point while
connecting P
new
to the trajectory T. Else, P
new
will be
connected to the last point L of the trajectory T,
normally, without any remove. Here, we spoke
about trajectories cleaning.
In case the MO has not been processed yet, the
algorithm creates a new trajectory T
new
for the new
MO and P
new
becomes the starting point of T
new.
This algorithm will stop runnning when all
existing points are processed. This is cheked by the
function check_point_occurence().
Figure 2: The trajectory construction algorithm.
Once the trajectory construction process is
achieved, the obtained trajectories are stored into the
TDB as a TDSrc.
As T1 is well performed, T2 takes place to
continue the whole transformation task. T2 focuses
on creating the target area models that consists of the
TDMs models generated directly from the relational
TDSrc model. This task translates the bottom up
approach features wherein the mapping process
plays an important role. Our goal is to identify how a
target field could be generated from the source field.
To this end, we have already defined a set of textual
rules in (Azaiez and Akaichi, 2015) that results the
generation of a set of TDM elements directly from
TDSrc at the aim to analyze mobile objects
trajectories obtained in T1. However, these textual
existing rules are inadequate to achieve this task
since their application in their state leads to waste
time and this is contradictory to our goal. Hence,
their translation basing on a transformation language
becomes a must. As a powerful mechanism, Model
Driven Engineering (MDE) is heavily building on
model transformations in converting models. It is a
general methodology for software development,
which recognizes the use of recurrent patterns to
increase compatibility between systems. Model
Driven Architecture (MDA) is a specific proposition
for implementing MDE. The MDA based
development of a software system starts by building
Platform Independent Models (PIM) of that system
which are refined and transformed into one or more
Platform Specific Models (PSMs) which can be used
to generate source code. MDA approach is
responsible for transforming the source model into
AlgorithmTrajectory construction
1. Point P
new
Extract_Point (x, y, t)
2. Do
3. If Trajectory.contains (MO) then
4. Point LTrajectory.lastPoint
5. If (check_point_state(P
new
,L,gap
space
,gap
time
)) then
6. Trajectory.add(P
new
)
7. If (check_liearity(L.LastPoint , L, P
new
)) then
8. Trajectory.remove (L)
9. End if
10. Else
11. Trajectory T
new
= new trajectory ()
12. T
new
.add(P
new
)
13. End if
14. Else
15. Trajectory T
new
= new trajectory ()
16. T
new
.add(P
new
)
17. End if
18.While (check_point_occurence())
Override Traditional Decision Support Systems - How Trajectory ELT Processes Modeling Improves Decision Making?
553
target models. It provides a set of guidelines for
structuring specifications expressed as UML models.
Each model is abstracted by its meta-model which
should be conformed to a meta-meta-model called
Meta Object Facility (MOF) (OMG, 2006).
Transformations have to be expressed by means of a
transformation language as ATLAS Transformation
Language (ATL) (ATLAS team, 2006). Figure 3
provides an overview of the ATL transformation
(Relational2Multidimensional) that generates the
multidimensional modeling, conforming to the meta-
model (MMTDM), from a TDSrc model that
conforms to its meta-model (MMTDSrc).
MMTDSrc, MMTDM and ATL metamodels are
expressed using the MOF semantics.
Figure 3: The MDA architecture.
The next figure presents the MMTDSrc.
Figure 4: Source meta-model.
Figure 5 describes the MMTDM.
Figure 5: Target meta-model.
In MDA approach, ATL specifications play an
important role to translate our proposed
transformation rules defined in (Azaiez and Akaichi,
2015) in order to generate the target model. An ATL
program is composed of a header, a set of side-
effect free functions called helpers and a set of rules
(Matched, called, lazy...) that are responsible to
perform transformations. Figure 6 presents a little
extract of the ATL code.
Figure 6: Relational2Multidimensional.atl.
Our work focuses on the convergence of all
TDMs that model organizational business events.
Once more than two TDMs are built, the fusion of
these brings about the creation of a central
repository of data named TDW (Figure 1).
4 CASE STUDY: EPILEPSY
PATIENT STATES
We propose to apply our approach to model a
Trajectory Decision Support System (TDSS) related
to persons suffered from epilepsy. This offers the
possibility to analyze the epilepsy patient state
evolution trajectory. Each year, thousands of
patients die following a seizure of epilepsy. In this
context, it is imperative that patients be able to
access the careers to contact in case of crisis. The
development of a new medical device called EpiLert
solves the problem. It comprises a wristband that is
based on a sensor capable of detecting the
movements of the limbs associated with an epileptic
convulsion to send automatically an alert signal to
caregivers when a person is in crisis while it
provides recordings of epileptic events and seizures
data in order to obtain additional medical tests. In
our case, extracting those data leads to collect a set
of raw TD. Once the loading process is achieved,
applying the trajectory construction algorithm
(Figure 2) becomes a must. Each crisis provides a
new trajectory belongs to such epilepsy patient. The
obtained trajectories of all patients are stored into
the TDB as a TDSrc model presented in the next
figure. This presents the T1 goal.
ColumnPrimary key
Table
Foreign key
1..* 0..*1..*
TDSrc model
1..*
1
0..*
0..*
0..*
0..*
0..*
TDM model
Table
Trajectory fact Dimension
Measure Trajectory fact identifier Dimension identifier Dimension proprety
Attribute
Hierarchy
1..*
0..*
1..*
1..*
1..*
1..*
Module Relational2Multidimensional;
Create out: Multidimensional from IN: Relational;
Rule S2S { /* Schema to Schema */
from s : IN!TDSrc_Schema
to t : OUT!TDM_Schema (Tables <- s.Tables)}
Rule T2TF { /* Table to Trajectory_Fact */
from s : IN!TDSrc_Table
to t : OUT!TDM_Trajectory_Fact(name <- ‘F’+’_’+s.name,
Measures <- s.Columns->collect(a|thisModule.C2M(a))
iskey<- s.iskey )}
Rule T2D { /* Table to Dimension */
from s : IN!TDSrc_Table
to t : OUT!TDM_Dimension (name <- ‘D’+’_’+s.name,
Attributes <- s.Columns->collect(e|thisModule.C2A(e))
iskey<- s.iskey )}
Lazy rule C2M { /* Column to Measure */
from s : IN!TDSrc_Column
to t : OUT! TDM_Measure(name <- s.name)}
Lazy rule C2M { /* Column to Attribute */
from s : IN!TDSrc_Column
to t : OUT! TDM_Attribute (name <- s.name,
iskey<- s.iskey )}
MODELSWARD 2017 - 5th International Conference on Model-Driven Engineering and Software Development
554
Figure 7: TDSrc model.
Since the TDSrc model is presented, it becomes
ready to be processed according to the second
transformation step T2 which goals at generating the
target models (TDM) from the TDSrc model basing
on the MDA approach. Transforming models using
MDA is related to the source model that must be
conform to its meta-model and the target meta-
model, recently defined, and the transformation rules
defined textually in (Azaiez and Akaichi, 2015) and
translated using ATL (Figure 6). TDSrc is a
relational model which consists of a set of tables and
columns that have to be transformed into a set of
multidimensional elements (facts, dimensions...).
According to our case study, applying the ATL
transformation rules leads to create a set of TDM
models. For example, the table Trajectory in the
TDSrc model feeds a trajectory fact table
F_Trajectory in the multidimensional design since it
satisfies conditions enumerated in the transformation
rule that is destined to identify trajectory facts. The
tables Patient, Move and Stop satisfy conditions
enumerated in the transformation rule that is
destined to identify dimensions. Therefore, they feed
respectively, D_Patient, D_Move and D_Stop. The
dimension D_Date is required in every
multidimensional schema since it contains all the
information we need about a certain date, and allows
analysts to analyze data as accurately as possible.
5 CONCLUSIONS
In this paper, we presented an overview on solutions
proposed by the research community to deal with the
ETL modeling problem. We expected the absence of
an ETL process that really leads to a better data
analysis. Hence, we relied on the Trajectory ELT
process to facilitate the propagation of the TD from
Operational Trajectory sources towards Trajectory
Warehouse area. Since the transformation task is the
core of the Trajectory ELT process, we proposed a
trajectory construction algorithm to transform raw
positions into trajectories and, therefore, generate
the TDSrc model. Then, we relied on the power of
the MDA mechanism to transform the obtained
source model into target models. We illustrated the
efficiency of our approach using a medical use case.
Currently, we are extending our framework to offer
a system which handles easily the evolution of the
whole warehousing chain while trajectory sources
evolve; this ensures reaching the BI goals, especially
extracting pertinent knowledge.
REFERENCES
ATLAS team, 2006. ATLAS Transformation Language
(ATL) Home page. http://www.eclipse.org/gmt/atl/.
Azaiez, N., Akaichi, J., 2015. How Trajectory Data
Modeling Improves Decision Making? In proceedings
of the 10th International Conference on Software
Engineering and Applications. Colmar, Alsace,
France, pp. 87-92.
Azaiez, N., Akaichi, J., 2016. What is the Impact of
Mobility Data Integration on Decision Support
Systems’ Modelling and Evolution? The International
Journal of Information Systems in the Service Sector
(IJISSS), 8(1), pp.1-12.
Boussaid, O. Bentayeb. F., Darmont, J., 2003. An MAS-
Based ETL Approach for Complex Data. In
proceedings of the 10th ISPE International
Conference on Concurrent Engineering: Research and
Applications. Madeira, Portugal, pp. 49-52.
Marketos, G., Frentzos, E., Ntoutsi, I., Pelekis, N.,
Raffaetà, A., & Theodoridis, Y., 2008. Building real-
world trajectory warehouses. In Proceedings of the
International ACM Workshop on Data Engineering
for Wireless and Mobile Access. Vancouver, BC,
Canada,pp. 8-15.
OMG, 2004. Model Driven Architecture (MDA).
http://www.omg.org/cgibin/doc?formal/03-06-01.
OMG, 2006. Meta Object Facility (MOF) 2.0 Core
Specification. OMG Document formal /2006-01-01.
Rifaieh, R., Benharkat, N. A., 2002. Query-based data
warehousing tool. In proceedings of the 5th ACM
International workshop on Data Warehousing and
OLAP, pp. 35 – 42.
Simitsis. A., 2003. Modeling and Managing ETL
Processes. VLDB Ph.D. Workshop.
Taktak, S., Alshomrani, S., Feki, J., Zurfluh, G., 2015. An
MDA Approach for the Evolution of Data
Warehouses. The International Journal of Decision
Support System technology (IJDSST); 7(3), pp. 65-89.
Vassiliadis, P., Simitsis, A., Georgantas P., Terrovitis, M.,
2003. A framework for the design of ETL scenarios.
In the proceedings of the 15th CAiSE. Velden, Austria,
pp 520-535.
Zekri, A., Akaichi, J., 2014. An ETL for Integrating
Trajectory Data A Medical Delegate Activities Use
Case Study. In the proceedings of the International
Conference on Automation, Control, Engineering and
Computer Science, pp.138-147.
Patient (id_pat, first_name_pat, last_name_pat, id_Epilert#, id_disease#)
Docteur(id_doc, first_name_doc, last_name_doc, id_service#)
Epilert(id_Epilert, color, id_device#)
Service Medical device (id_device, marque)
Hospital service (id_sevice, designation)
Disease(id_disease, name_disease)
Trajectory (id_traj, #id_move, #id_stop, #id_pat)
Stop (id_stop, #id_begin, #id_end)
Move (id_move, #id_begin, #id_end)
Begin(id_begin, Begin_time)
End (id_end, End_time)
Override Traditional Decision Support Systems - How Trajectory ELT Processes Modeling Improves Decision Making?
555