Hybrid of Regression Trees & Linear Regression algorithm was initially developed for such cases, where there are no perfect linear dependence between features and responses, but, however, there are indicators of some linear relationships between them. In such cases basic linear regression algorithm, which searches for best linear model, may fail. Indeed, we can recall "Simpson's paradox", according to which combining two sets with similar trends may lead this trend to turn into opposite or totally disappear. On the other hand, algorithms, that ignore linear relationships at all, could not achieve the best quality. HRT algorithm helps to solve both issues.
In our project we reviewed main literature about HRT algorithm, implemented it and aЗplied for several datasets: PIK, StarSkill and Airfoil.