Towards forest fire prediction in Algeria

Abstract

This work was carried out as part of our end-of-cycle master project in computer science Software Engineering option. It consisted of a machine learning approach to the prediction of forest fres in Algeria. Predictive modeling is an emerging feld, in recent years several researches are interested in the task of predictive analysis, especially in the forestry feld. Our work focuses on predicting forest fres in the regions of Algeria. The Dataset we used was taken from the Kaggle website containing fre data from two regions in Algeria : Béjaïa and Sidi-Bel-Abbes. The prediction was made on the meteorological data and then classifed using an algorithm of the Machine Learning. Machine learning is the most practical technique. So it would be much more easy to predict the possibility of a wildfre if a model was adopted to polarize them and learn from them. In this research work, forest fre data were analyzed and classifed by logistic regression. Our brief consists of six (6) chapters organized as follows : In the frst chapter, we defned our context and problematic as well as our objectives, we have also detailed our working methodology. In the second chapter, we presented some defnitions of the domain studies that is predictive modeling, its types, problems related to this feld and Machine Learning and its types. In the third chapter, we have established the state of the art that represents all the related works that we have synthesized, we have presented this in a table which contains the outline of each synthesized approach, while following each work by a brief paragraph that summarizes it, then we proceeded to an analysis comparative between the approaches of the related documents and our approach. In the fourth chapter, we presented in detail the approach we used during our project as well as its di?erent steps to make a prediction of fres from datasets. In the ffth chapter, we have discussed the various aspects related to the implementation of the approach we have developed, namely, technologies, software and the languages chosen using di?erent data sources for implementation of our approach. Despite the difculties we had during the realization of our work, such as the lack of databases, the lack of sources of information, we pushed the project as far as possible ; there are still many steps to add. In particular the implementation of some steps of our approach, namely, We are thinking of refning our approach and implementing it in better hardware and software conditions. Replace the logistic regression algorithms with other algorithms in order to improve the accuracy of the results obtained and build a more efcient and e?ective prediction system. Finally, apply the model really proposed on the ground and its exploitation by the Directorate of the Conservation of Forests.