Pseudo dinucleotide composition (PseDNC) and Z curve showed excellent performance in the classification issues of nucleotide sequences in bioinformatics. Inspired by the principle of Z curve theory, we improved PseDNC to give the phase-specific PseDNC (psPseDNC). In this study, we used the prediction of recombination spots as a case to illustrate the capability of psPseDNC and also PseDNC fused with Z curve theory based on a novel machine learning method named large margin distribution machine (LDM). We verified that combining the two widely used approaches could generate better performance compared to only using PseDNC with a support vector machine based (SVM-based) model. The best Mathew's correlation coefficient (MCC) achieved by our LDM-based model was 0.7037 through the rigorous jackknife test and improved by ∼6.6%, ∼3.2%, and ∼2.4% compared with three previous studies. Similarly, the accuracy was improved by 3.2% compared with our previous iRSpot-PseDNC web server through an independent data test. These results demonstrate that the joint use of PseDNC and Z curve enhances performance and can extract more information from a biological sequence. To facilitate research in this area, we constructed a user-friendly web server for predicting hot/cold spots, HcsPredictor, which can be freely accessed from . In summary, we provided a united algorithm by integrating Z curve with PseDNC. We hope this united algorithm could be extended to other classification issues in DNA elements.