Sparse regression with Multi-type Regularized Feature modeling

Within the statistical and machine learning literature, regularization techniques are often used to construct sparse (predictive) models. Most strategies only work for data where all predictors treated identically, such as Lasso regression (continuous) linear effects. However, many predictive problems involve different types of require a tailored term. We propose multi-type penalty that acts on objective function sum subpenalties, one each type predictor. As such, we allow predictor selection level fusion within in data-driven way, simultaneous with parameter estimation process. develop new strategy convex models this penalty. Using theory proximal operators, our procedure is computationally efficient, partitioning overall optimization problem into easier solve subproblems, specific its associated Earlier research applies approximations non-differentiable penalties problem. The proposed SMuRF algorithm removes need achieves higher accuracy computational efficiency. This demonstrated an extensive simulation study analysis case-study insurance pricing analytics.