Gere previsões ortogonais (não correlacionadas) para uma determinada variável

Eu tenho uma Xmatriz, uma yvariável e outra variável ORTHO_VAR. Preciso prever a yvariável usando X, no entanto, as previsões desse modelo precisam ser ortogonais e, ao ORTHO_VARmesmo tempo, estar o mais correlacionadas ypossível. Eu preferiria que as previsões fossem geradas com um método não...

correlation machine-learning dataset logistic-regression prediction linear-regression prediction dummy-variables neural-network image-classification python k-nn python neural-network neural-network deep-learning keras tensorflow image-classification tensorflow reinforcement-learning policy-gradients machine-learning decision-trees neural-network overfitting data-analysis metric python scikit-learn distance scipy machine-learning python scikit-learn decision-trees logistic-regression keras image-classification implementation machine-learning python scikit-learn random-forest decision-trees machine-learning feature-selection feature-engineering word2vec word-embeddings natural-language-process scikit-learn time-series clustering k-means python cross-validation pyspark statistics cross-validation multiclass-classification evaluation machine-learning nlp machine-translation neural-network deep-learning keras tensorflow image-classification machine-learning python similarity distance lstm text named-entity-recognition machine-learning keras optimization gan learning-rate neural-network data-mining dataset databases books neural-network rnn