<p>Avalanche forecasting is most widely practiced approach for avalanche hazard mitigation in snow bound mountainous regions. In this context, machine learning techniques are typically used to discriminate between avalanche or no-avalanche scenarios. However, limited studies have been focused on selection of significant features that influence avalanche formation. Consequently, prevailing forecasting models often access redundant data leading to slower learning processes, increased computational complexity, and potential overfitting that ultimately compromises its generalization ability. This study proposes adoption of feature selection as a fundamental step in avalanche forecast modeling. In this regard, we demonstrate the usage of a popular feature selection technique SVM-RFE to identify the most relevant feature subset for avalanche forecasting in Bandipore-Gurez region in North-western part of Indian Himalayas. The study explored a set of 40 features (including the features that represented snow-meteorological perturbations occurred over past many days) and identified subsets of 7 to 15 features representing fresh rainfall, fresh snow, cumulative seasonal snow, temperatures, wind speed and direction, and sunshine as the most significant contributors to the avalanche occurrences. Also, some of these selected features represent information up to past 2 to 3 days indicating that the avalanching conditions in the study area ripe over a longer period of time rather than resulting from rapid meteorological changes. Further, the performances of classification models with selected subsets of features either matched or exceeded their respective performances with entire set of 40 features. These findings provide valuable insights for researchers and practitioners in avalanche hazard management and may contribute to development of more efficient forecasting models.</p>