PROJECT REPORT FORECASTING DATABASE STORAGE USING MACHINE LEARNING

TANTIO, CHRISTOPHORUS REYHAN (2019) PROJECT REPORT FORECASTING DATABASE STORAGE USING MACHINE LEARNING. Other thesis, UNIKA SOEGIJAPRANATA SEMARANG.

[img]
Preview
Text (COVER)
16.K1.0028 CHRISTOPORUS REYHAN_COVER_a.pdf

Download (1MB) | Preview
[img]
Preview
Text (BAB I)
16.K1.0028 CHRISTOPORUS REYHAN_BAB I_a.pdf

Download (1MB) | Preview
[img] Text (BAB II)
16.K1.0028 CHRISTOPORUS REYHAN_BAB II_a.pdf
Restricted to Registered users only

Download (2MB)
[img]
Preview
Text (BAB III)
16.K1.0028 CHRISTOPORUS REYHAN_BAB III_a.pdf

Download (1MB) | Preview
[img]
Preview
Text (BAB IV)
16.K1.0028 CHRISTOPORUS REYHAN_BAB IV_a.pdf

Download (2MB) | Preview
[img]
Preview
Text (BAB V)
16.K1.0028 CHRISTOPORUS REYHAN_BAB V_a.pdf

Download (2MB) | Preview
[img]
Preview
Text (BAB VI)
16.K1.0028 CHRISTOPORUS REYHAN_BAB VI_a.pdf

Download (1MB) | Preview
[img]
Preview
Text (DAFTAR PUSTAKA)
16.K1.0028 CHRISTOPORUS REYHAN_DAPUS_a.pdf

Download (1MB) | Preview
[img]
Preview
Text (LAMPIRAN)
16.K1.0028 CHRISTOPORUS REYHAN_LAMPIRAN_a.pdf

Download (3MB) | Preview

Abstract

Space/storage management in database rely on the experience of Database Administrator (DBA) to predict time left until full and supported by script to alert when the database is full using precentage. There’s no other way Database Administrator to predict days left until database is full without experience and feeling. For newly created database, Database Administrator can’t predict when the database is being full in second, third, until twelve month (depends on how experienced the DBA is). The precentage threshold that is used in alert system also provide a problem because if the started size is 100MB then 10% of free space is 10MB, but if the size reached 1000MB, 10% of it is 100MB which is still a lot of free space. This study trying to solve that problem by applying Machine Learning approach. This research use subset of machine learning, Deep Neural Network that use multiple layer of neurons in Neural Network to extract useful data from a pattern in the data. Deep Neural Network algorithms get implemented in the database system to forecasts time in days until the database considered full. The algorithm use simulation of oracle production database based on a retail company as corpus. Size and other various of data that are fetched from oracle database about a tablespace are processed so it can be used to get the best input parameters for training and predicting. The processed data that has output will be divided to training, validating, and testing, with the training and validating use most of the dataset. This testing used for monitoring the performance of Deep Neural Network and change the neural network settings and input to get the best performance based on the accuracy(what considered accurate are when the prediction is the same days as the real data or the prediction only has one day difference from the real data) . This system uses precentage and bytes as threshold that function as a flag if the current database size is considered full. When it’s full, this system use the Deep Neural Network to train using past data to predict the future data. Supervised method is used with days until current data is full as output. The result from this research is that Deep Neural Network can be used for prediction on days left until oracle database tablespace is full to some degree. Each cycle could have different pattern in adding extent and that makes the accuracy goes down. The most optimal structure is using 3 input neurons, 3 hidden layer with shape of diamond (2 x total input – total input + (total input/2) – total input), and 1 output neurons.

Item Type: Thesis (Other)
Subjects: 000 Computer Science, Information and General Works > 004 Data processing & computer science
Divisions: Faculty of Computer Science > Department of Informatics Engineering
Depositing User: Ms Agustin Hesti Pertiwi
Date Deposited: 23 Jun 2020 04:15
Last Modified: 23 Sep 2020 02:01
URI: http://repository.unika.ac.id/id/eprint/21440

Actions (login required)

View Item View Item