Bhagat, NarayanKumar and Singh, Rakesh K and Sawmliana, C. and Singh, P.K. (2022) Application of logistic regression, CART and random forest techniques in prediction of blast-induced slope failure during reconstruction of railway rock-cut slopes. Engineering Failure Analysis, 137.

Full text not available from this repository. (Request a copy)

Abstract

Drilling and blasting operation is often required to excavate the infrastructure slopes for enhancing their stability or creating space for upgradation. While conducting blasting, there are many incidents of slope failure or rockfall. Thus, proper planning and careful designing of different blasting parameters are essentially required to reduce the incidents of slope failure or rockfall. In the present research, the efficacies of three machine learning (ML) techniques; Logistic Regression (LR), Classification and Regression Tree (CART) and Random Forest (RF) were examined for predicting the blast-induced slope failure (BISF) or blast-induced rockfall during reconstruction of slopes on railway route. 490 databases with thirteen variables were considered for the prediction of BISF. By applying Multicollinearity and LR technique based on minimum Akaike Information Criterion values, the six most influential input parameters were identified. With the selected input datasets, fivefold cross-validation was carried out on randomly selected five sub-groups of datasets using LR tool. Then, the best LR model having the highest prediction rate was selected and with the same training and testing datasets of the selected model, the CART and RF models were also developed. The various performance indices such as correctness, recall rate, precision, specificity, F-beta score, receiver operating characteristics (ROC) and area under the curve (AUC) were calculated to evaluate the developed models' accuracy and applicability. The developed models showed good prediction abilities, with the RF model having highest performance in terms of recall rate (90%), accuracy (96.94%) and F-beta score (0.882). The LR model has higher precision (88.9%) and AUC value (0.96) than CART and RF models. The findings of the research work demonstrate the applicability of all three models in selecting the blast design parameters to prevent BISF during blasting. The use of developed models would result in saving the commuter’s lives, avoiding traffic delays and minimising property damages in similar situations.

Item Type:	Article
Uncontrolled Keywords:	Railway rock-cut-slopeBlast design parametersBlast-induced slope failureMachine learning technique
Subjects:	Blasting
Divisions:	UNSPECIFIED
Depositing User:	Mr. B. R. Panduranga
Date Deposited:	28 Mar 2022 04:10
Last Modified:	28 Mar 2022 04:10
URI:	http://cimfr.csircentral.net/id/eprint/2489

Actions (login required)

View Item