Skip to Main content Skip to Navigation
New interface
Theses

Data management in forecasting systems : optimization and maintenance

Abstract : Forecasting systems are usually based on data warehouses for data strorage, and OLAP tools for historical and predictive data visualization. Aggregated predictive data could be modified. Hence, the research issue can be described as the propagation of an aggregate-based modification in hirarchies and dimensions in a data warehouse enironment. Ther exists a great number of research works on related view maintenance problems. However, to our knowledge, the impact of interactive aggregate modifications on raw data was not investigated. This CIFRE thesis is supported by ANRT and the company Anticipeo. The application of Anticipeo is a sales forecasting system that predicts future sales in order to draw appropriate business strategy in advance. By the beginning of the thesis, the customers of Anticipeo were satisfied the precision of the prediction results, but not with the response time. The work of this thesis can be generalized into two parts. The first part consists in au audit on the existing application. We proposed a methodology relying on different technical solutions. It concerns the propagation of an aggregate-based modification in a data warehouse. the second part of our work consists in the proposition of a newx allgorithms (PAM - Propagation of Aggregated-baseed Modification) with an extended version (PAM II) to efficiently propagate in aggregate-based modification. The algorithms identify and update the exact sets of source data anf other aggregated impacted by the aggregated modification. The optimized PAM II version archieves better performance compared to PAM when the use of additional semantics (e.g. dependencies) is possible. The experiments on real data of Anticipeo proved that the PAM algorithm and its extension bring better perfiormance when a backward propagation.
Document type :
Theses
Complete list of metadata

Cited literature [60 references]  Display  Hide  Download

https://theses.hal.science/tel-00997235
Contributor : ABES STAR :  Contact
Submitted on : Tuesday, May 27, 2014 - 4:52:07 PM
Last modification on : Friday, September 30, 2022 - 11:34:15 AM
Long-term archiving on: : Wednesday, August 27, 2014 - 12:05:26 PM

File

TH2012_Feng_Haitang_ii.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-00997235, version 1

Citation

Haitang Feng. Data management in forecasting systems : optimization and maintenance. Other [cs.OH]. Université Claude Bernard - Lyon I, 2012. English. ⟨NNT : 2012LYO10174⟩. ⟨tel-00997235⟩

Share

Metrics

Record views

394

Files downloads

391