A comparative evaluation of supervised machine learning algorithms for township level landslide susceptibility zonation in parts of Indian Himalayas

•Township level LSZ using novel ELANFIS technique attempted.•Comparative appraisal of ELANFIS vis-à-vis ANN, ELM and SVM techniques.•Evaluation of reliability & prediction capability of LSZ models using FR and ROC.•Replicable for future sustainable urbanization in line with DRR goals. Landslide susceptibility zonation (LSZ) has generally been regarded as the appropriate stride to begin scientific studies in mountainous terrains to alleviate the socio-economic consequences of landslides. Application of machine learning (ML) with geographic information system (GIS) is a promising fusion of technologies, for spatial prediction of landslide susceptibility with high precision, and has been applied widely in the past. However, the literatures of ML and GIS-based LSZ gives a fuzzy conclusion upon the righteous choice of ML technique among many state-of-the-art techniques, and do not present a probe on the aptitude of ML models for township level LSZ attempts. This research investigates such concern with a case study to figure out a robust technique, which can be a benchmark approach in future case studies and various comparisons strives across the sundry genre of ML. For that, the present attempt has been anchored to four different supervised ML algorithms including artificial neural network (ANN), extreme learning machine (ELM) of neural network (NN) genre, classical ML algorithm of support vector machine (SVM) and extreme learning adaptive neuro fuzzy inference system (ELANFIS) of neuro-fuzzy system genre. The Mussoorie Township, a famed hill station in the Indian State of Uttarakhand was chosen as the area for case study. A total of 13 landslide susceptibility maps (LSM) were produced. Spatial performance of these maps was compared and statistically validated with the help of landslide inventory of the study area. Amongst the LSMs, the LSM-ELANFIS-VII of ELANFIS model with 11 number of membership functions (MF) was found to be in better agreement with all the validation measures performed. In addition to the satisfactory performance on validation, the LSMs produced through ELANFIS display a unique trace of geomorphological features on it along with pragmatic scattering of landslide susceptibility classes - an omen that exhorts graduation of GIS-based LSZ to ensemble neuro-fuzzy ML models.