Extracting Features from Textual Data in Class Imbalance Problems

We address class imbalance problems. These are classification problems where the target variable is binary, and one dominates over other. A central objective in these to identify features that yield models with high precision/recall values, standard yardsticks for assessing such models. Our extracted from textual data inherent use n-gram frequencies as introduce a discrepancy score measures efficacy of an highlighting minority class. The frequency counts n-grams highest scores used construct desired metrics. According best practices followed by services industry, many customer support tickets will get audited tagged “contract-compliant” whereas some be “over-delivered”. Based on in-field data, we random forest classifier perform randomized grid search model hyperparameters. scoring performed using function. minimize follow-up costs optimizing recall while maintaining base-level precision score. final optimized achieves acceptable staying above precision. validate our feature selection method comparing constructed chosen randomly. propose extensions extraction general (binary multi-class) regression measure dissimilarity distributions other (more general) formulate could potentially more effective