The invention provides a heuristic algorithm for credit assessment feature selection, and the algorithm generates a random spider population based on an SSA algorithm, calculates an opposite solution of the spider population by employing an OBL strategy, selects an optimal solution to form the number of OBL populations, carries out the algorithm iteration of the randomly generated spider population and the OBL spider population, calculates the fitness value and vibration value of a spider individual, selects an optimal solution individual by using a local search algorithm LSA, enables unselected individuals to enter a next round of iteration, and outputs all optimal solutions selected by the LSA after the iteration is finished to form an optimal solution set. According to the invention, the algorithm in the invention is learned through a machine instead of traditional artificial feature screening, so that the efficiency of feature screening is improved; compared with a general heuristic algorithm, an OBL strategy is added into the algorithm, so that the space coverage rate and the stability of the algorithm are remarkably improved; according to the algorithm, an LSA algorithm architecture suitable for the P2P field is introduced, and the feature screening accuracy and the model matching degree are improved.