Contribution de (Feature Engineering) à l’amélioration des Modèles de la Régression Linéaire Multiple pour la Prédiction du Carbone Organique Totale, Application aux Schistes de champs de Barnett (Etats Unis)

Abstract

Total organic carbon TOC values the richness of the source rock in organic matter and its predisposition to generate hydrocarbons. The TOC content qualifies the potential of the source rock. The best source of TOC estimation is the direct measurement from the cores in the laboratory, but because of the geological or technical problems during the coring operation the recovery of the cores is not always integral, In addition, the dosage TOC is expensive and time consuming.In our work on Barnett's shale gas field, multiple regression and Feature Engineering and it’s models were applied on classical logs (Gamma ray, Resistivity, Sonic and Density) to improve the results, they gives great precision (CC=0.88 ; 0.95, RMSE=0.54;0.06, MAE=0.40 ; 0.05(,Thanks to the LOG10(DATA) without OUTLIERS which has a great performance, we can solve the problem of 51 unmeasured points, minimize cost and time saving..