مسعود جان بك لو

Detecting patients in the early stages of the disease can prevent the spread of the epidemic in the community and save lives. On the other hand, the capacity of the standard corona detection method (RT PCR test) is limited and it took one to two days to determine the result. The purpose of this research was to develop diagnosis and prognosis models with description based on value blood counts of patient using machine learning. To achive this purpose 598 samples was used as dataset . Five machine learning algorithms were eva‎luated on this dataset : Logistic regression, support vector machine, random forest, gradient boosting tree and XGBOOST. Bayesian optimization method along with five-step cross-validation was used to set the hyperparameters of the algorithms. Two eva‎luation criteria were used to select the best model for the diagnosis and prognosis phase : The performance of models in five stages of cross validation and the performance of models in predicting the test set. Also backtest performance (Detecting negative covid-19 patient but positive for others respiratory pathologies) for disgnosis phase . Feature importance and SHAP analysis was used to explain chosen models. random forest model with f1 score 77 % , AUC score 85.9% , sensivity 74.4% and specificity 97.5 % had the best results in diangnosis models . Gradient Boosting Tree model with f1 score 77% < AUC score 79 % , precision 76 % and recall 80 % had the best results in prognosis models . important variables for diagnosis model were leckocyte , eosinophil and platelets. Age and lymphocyte and leukocyte were important in prognosis model.Diagnosis model was able to obtain better results in terms of sensitivity and specificity criteria compared to studies that used the same data set. Also, this research highlighted the importance of describing the predictive model to show the blood variables that are possible in the diagnosis and prognosis. The structure of the research method used in this research can be used for the diagnosis and prognosis of other diseases.

بيماري كرونا , تشخيص , پيش آگهي , يادگيري ماشين

covid 19 , diagnosis , prognosis , machine learning

masood janbackloo

alireza aliahmadi