Enabling Effective Data Mining of RFID Data

8/8/2019 Enabling Effective Data Mining of RFID Data

1/28

Roma Chauhan Rajeev Gupta

Computer Science DepartmentInstitute of Management Education

Sahibabad

8/8/2019 Enabling Effective Data Mining of RFID Data

2/28

What we going to cover

y Introduction to RFID System

y Applying RFID: Supply Chain

y RFID Data Characteristics and issues

y Why RFID Data Mining?

y RFID Data Mining

y RFID Data Mining Challenges

y Conclusions

8/8/2019 Enabling Effective Data Mining of RFID Data

3/28

B ar Code

8/8/2019 Enabling Effective Data Mining of RFID Data

4/28

B arcode Disadvantagesy In order to keep up with inventories, companies must

scan each bar code on every box of a particularproduct.

y Going through the checkout line involves the sameprocess of scanning each bar code on each item.

y Bar code is a read-only technology, meaning that itcannot send out any information.

8/8/2019 Enabling Effective Data Mining of RFID Data

5/28

RFI Dy RFID is an technology that uses radio-frequency waves

to transfer data between a reader and a movable item

to identify, categorize, track...y RFID is fast, reliable, and does not require physical

sight or contact between reader/scanner and thetagged item

y It uses technologies using radio wave to automatically identify objects

8/8/2019 Enabling Effective Data Mining of RFID Data

6/28

y RFID tags are intelligent bar codes that can talk to anetworked system to track every product that you put

in your shopping cart.y Many manufacturers use the tags to track the location

of each product they make from the time it's madeuntil it's pulled off the shelf and tossed in a shopping

cart.

8/8/2019 Enabling Effective Data Mining of RFID Data

7/28

Why RFI Dy Read/Writey A bility to add information directly to tags enables each

unique asset to carry its own unique history y N on-contact Readsy A

bility to read tags at a distance, under a variety of environmental conditions, without physical manipulation of the asset

y Fast Ready

A bility to simultaneously read large numbers (1000-1750tags/sec) of itemsy Automation: Requires less human interventiony Authenticity y Each RFID chip is unique and can not be replicated

8/8/2019 Enabling Effective Data Mining of RFID Data

8/28

8/8/2019 Enabling Effective Data Mining of RFID Data

9/28

Components of an RFID Systemy Tag: The tag contains a microchip for storage of data and an antenna

which is tuned to receive radio frequency waves emitted by a scannerfor allowing wireless transmission of data to the reader.

y Scanner : It is a reader which contains radio frequency module, acontrol unit and a coupling element to interrogate the tags via radiofrequency communication. Readers are usually connected throughmiddleware to a back-end database.

y The Middleware which interacts with backend database. Middleware isresponsible for cleaning the data such as eliminating false reads besidesperforming aggregation and filtering of data. A lso, by monitoringmultiple readers, middleware can detect the movement of RFID tags asthey pass from the read range of one reader to another,

8/8/2019 Enabling Effective Data Mining of RFID Data

10/28

RFI D Technologyy Radio Frequency Identification (RFID)

y Technology that allows a sensor (reader) to read,from a distance, and without line of sight, a uniqueelectronic product code (EPC) associated with a tag

8/8/2019 Enabling Effective Data Mining of RFID Data

11/28

RFI D System (Tag, R eader, Database)

8/8/2019 Enabling Effective Data Mining of RFID Data

12/28

Electronic Product Code (EPC)y

EPC is Coding schemes created as an eventual successor to the barcode.y The EPC was created as a low-cost method of tracking goods using RFIDtechnology. It is a standard naming scheme by A uto-Id Center.

8/8/2019 Enabling Effective Data Mining of RFID Data

13/28

Events during RFI D tracking

8/8/2019 Enabling Effective Data Mining of RFID Data

14/28

App lying RFI Dy RFID Enabled Supply Chain: Inventory management, reduce thrift,

easy and effective tracking of items.y Retail : A ctive shelves monitor product availability y Access control : Toll collection, credit cards, building accessy Airline luggage management : Reduce lost/misplaced luggagey Medical : Implant patients with a tag that contains their medical

history y P et identification : Implant RFID tag with pet owner informationy

8/8/2019 Enabling Effective Data Mining of RFID Data

15/28

M ovement of an object within a su pp ly chain

before it is finally p urchased by the customer

8/8/2019 Enabling Effective Data Mining of RFID Data

16/28

RFI D Data Characteristics and issuesy The RFID datasets possess following set of characteristics:

y Inaccuracy y Massive data streamsy Temporal & spatial featuresy Integrationy Scalable Systemsy Heterogeneous datasetsy Inaccuracy y Inferencesy Manageabley Security y Mapping physical and interpreted world

8/8/2019 Enabling Effective Data Mining of RFID Data

17/28

Inaccuracy y The RFID data is dirty and not very accurate due to duplicate

and missed reads by the reader.y The RFID data seems to be noisy and unreliable in a raw

structure due to false reads.

Massive data streamsy RFID datasets are continuously generated stream of data

recorded all the time by RFID readers automatically.

y Thus huge data requires scalable storage schemes for efficientstorage of RFID data.

8/8/2019 Enabling Effective Data Mining of RFID Data

18/28

Temporal & spatial featuresy During the reading of the RFID datasets it is depended on

time stamps (temporal). A lso, the RFID datasets are spatial innature as they move across locations during their life cycle.This add to another layer of complexity

Integrationy Integrating of RFID data with company s backend operations

& business processes many challenges.y The cost of implementing RFID infrastructure within the

boundaries of an organization bears a large cost for thecompany.

8/8/2019 Enabling Effective Data Mining of RFID Data

19/28

Scalable Systemsy With the enormous growth of RFID databases the system

should be scalable enough to handle it.y The system should be capable enough to grow in one or more

dimensions with the increase in the volume of RFID data andthe number of transactions without affecting performance.

Heterogeneous datasetsy The Organizations that adopt RFID technology must handle

data from thousands of readers distributed across variousdispersed geographical locations. The RFID solution can bedeployed across multiple sites, companies, or even countries.

8/8/2019 Enabling Effective Data Mining of RFID Data

20/28

y Inferencesy The RFID datasets always carries embedded implicit information

such as changes of state and containment relationships amongobjects. To make proper inferences of RFID datasets it requirescontext of other information also.

y Manageabley To manage huge datasets good support of administration and

testing is a prerequisite for the successful deployment of an RFIDsolution in large-scale, distributed applications.

y Security y The vast amount of potentially sensitive information involved in

RFID systems makes security concerns critical. Such like:placement of wrong RFID tags on the objects.

8/8/2019 Enabling Effective Data Mining of RFID Data

21/28

RFI D Data M ining: Whyy The massive RFID data can be mined to produce

prediction analysis and determine patterns in realtime that can help further to improve supply chainprocess

y Tracking of objects moving in the supply chaincontains valuable knowledge important and very much required to understand processes, such asinventory management, and quality control incomplex production systems.

y Vehicle tracking data can be used by transportation managers for incident detection orroad network planning.

8/8/2019 Enabling Effective Data Mining of RFID Data

22/28

y The massive RFID data brings great opportunity formining techniques.

y

The huge datasets can be mined to produce predictionanalysis and determine patterns in real time.

8/8/2019 Enabling Effective Data Mining of RFID Data

23/28

Issues in Data Cleaningy L ack of Completeness

y RFID readers capture only 60-70% of all tags that are inthe vicinity

y Smoothing of data is done to rectify the loss of intermediate messages

y Temporal Nature of data or tag dynamicsy RFID tags are in motion and that is what makes them

more difficult to handley But motion of a tag causes dropping of messages

y RFID data streams are very fast and are huge innumber

y Hence filtering is important before sending them to

8/8/2019 Enabling Effective Data Mining of RFID Data

24/28

Challenge for RFI D data miningy The major problem with RFID data is that the volume of

data increases as product moves in supply-chain in terms of time and location. Wal-Mart is expected to generate 7terabytes of RFID data per day

y Moreover, data is redundant that needs to be consolidatedand transformed to occupy less space in the database.

y To obtain the desired result one must ensure that no usefulinformation is lost during the process.

8/8/2019 Enabling Effective Data Mining of RFID Data

25/28

Characteristic of RFI D Datay Spatio-temporal, dynamics and correlationsy Data contains time, location, and statusy Rich semantics.y Objects carry a lot of informationy related with its context status and background

knowledgey U ncertainty and heterogeneity y

missing readings, and repeat readingsy dirty datay Streaming, batching and massive volumey automatically generated rapidly in form of streamingy Objects must be checked in a batch

8/8/2019 Enabling Effective Data Mining of RFID Data

26/28

Challengesy The gigantic size of RFID data, and the diversity of

queries poses great challenges to traditional system

and analysis technology since processing may involveretrieval and reasoning over a large number of interrelated tuples through different stages of objectmovements

8/8/2019 Enabling Effective Data Mining of RFID Data

27/28

Data Cube Challengesy The path nature of RFID data makes it hard to incorporate

into a traditional data cube while preserving its structurey We can count the number of objects that stayed at a

particular location for defined period of time. Such as if we want to calculate the number of objects that stayed at aparticular location for what duration of time. But, it sdifficult to determine distance object covers while movingfrom one store location to the other store location.

y The cleaned RFID data moves beyond raw tuples form. Itdepends on path database which is (code, time in, timeout).

8/8/2019 Enabling Effective Data Mining of RFID Data

28/28

Thank You.