Learning analytics (LA) is a significant field of study to examine and identify difficulties the novice programmers face while learning how to program. Despite producing notable research by the community in the specified area, rare work is observed to synthesize these research efforts and discover the dimensions that guide the future research of learning analytics in programming courses (LAPC). This work demonstrates review of the learning analytics research for initial level programming courses by exploring different types and sources of data used for LA, and evaluating some pertinent facets of reporting, prediction, intervention, and refinements exhibited in literature. Based on the reviewed aspects, a taxonomy of LAPC research has been proposed along with the associated benefits. The results reveal that most of the learning analytics studies in programming courses used assessment data, which is generated from conventional assessment processes. However, the analysis based on more granular level data covering the cognitive dimensions and concept specific facets could improve accuracies and reveal the precise aspects of learning. In addition, the coding analysis parameters can broadly be categorized into code quality and coding process. These categories can further be classified to present twenty-five sub-categories of coding parameters for analyzing the behaviors of novice programmers. Moreover, efforts are required for early identification of effective and ineffective behavioral patterns through performance predictions in order to deliver timely interventions. Lastly, this review emphasizes the integration of related processes to optimize the future research efforts of conducting the learning analytics research for programming courses.