| Titre : | In Fulfillment of the Requirements for the Degree | | Type de document : | projet fin études | | Auteurs : | ARFAOUI Hamza, Auteur | | Langues : | Français (fre) | | Catégories : | e-Logistique
| | Index. décimale : | 2160/19 | | Résumé : | In order to predict the future and take the adequate decision to influence it, we use
technology in predictive analytics. Organizations can use historical performance data to
extrapolate and make predictions about the future and take actions that would affect those
results.
Concept drift is the main cause to decrease the accuracy of data predictions on
certain fields as time passes. And to remedy this problem, we can adopt both passive and
active solutions. Like change detection tests to detect concept drift as a change in the
timeseries. When a drift is detected, we need to update immediately the model to maintain
its accuracy.
The goal of this thesis is to implement a reliable drift detection tool so that the
company –SimAnalytics- can guarantee the accuracy of its prediction models. The
purpose of the case study is to benchmark the most used drift detection mathematical
models, understand their principle, then implement them in python. Only then we can test
the modules on our datasets and decide which method is reliable and fast for our test
cases.
This internship took place at the company’s headquarter in Helsinki, Finland from
the 1st of June 2019 to the end of September of the same year.
This work can be divided in four main phases described as follows:
• Phase I: Introduction
• Phase II: Preliminary study and project description
• Phase III: Code Implementation and data generation
• Phase IV: Experimental results and results analysis |
In Fulfillment of the Requirements for the Degree [projet fin études] / ARFAOUI Hamza, Auteur . - [s.d.]. Langues : Français ( fre) | Catégories : | e-Logistique
| | Index. décimale : | 2160/19 | | Résumé : | In order to predict the future and take the adequate decision to influence it, we use
technology in predictive analytics. Organizations can use historical performance data to
extrapolate and make predictions about the future and take actions that would affect those
results.
Concept drift is the main cause to decrease the accuracy of data predictions on
certain fields as time passes. And to remedy this problem, we can adopt both passive and
active solutions. Like change detection tests to detect concept drift as a change in the
timeseries. When a drift is detected, we need to update immediately the model to maintain
its accuracy.
The goal of this thesis is to implement a reliable drift detection tool so that the
company –SimAnalytics- can guarantee the accuracy of its prediction models. The
purpose of the case study is to benchmark the most used drift detection mathematical
models, understand their principle, then implement them in python. Only then we can test
the modules on our datasets and decide which method is reliable and fast for our test
cases.
This internship took place at the company’s headquarter in Helsinki, Finland from
the 1st of June 2019 to the end of September of the same year.
This work can be divided in four main phases described as follows:
• Phase I: Introduction
• Phase II: Preliminary study and project description
• Phase III: Code Implementation and data generation
• Phase IV: Experimental results and results analysis |
|
216/01 
Faire une suggestion Affiner la recherche
In Fulfillment of the Requirements for the Degree / ARFAOUI Hamza
