Risk prediction of household mite infestation based on machine learning

Abstract House dust mites produce the allergens causing allergic diseases. Predicting risk level of mites infestation through environmental conditions instead of using complex detection methods can increasing people's attention to mites and avoid huge detection cost. Mite allergens Der 1 (Der f 1 + Der p 1) in 101 residential apartments in different regions of China were measured. Indoor environmental parameters were continuously monitored and occupant surveys were also collected. Der 1 over 2000 ng/g is defined as mite infestation risk level. Compared with logistic regression and support vector machine (SVM), the prediction results of the extremely gradient boosting (XGBoost) model are more accurate (ACC = 0.838, Recall = 0.844) and highly interpretable, thus it is the most suitable method for mite risk level prediction. Indoor environment data of other cities in China were collected for prediction, “risk level map of mite infestation in China” was produced. The predicted results showed that risk level of mite infestation in southern China are generally higher than in northern cities. Based on the well-trained XGBoost prediction model, the relationship between input features and probability of mite infestation reaching risk level (PR=1) can be found: 1) “mite active breeding zone” was defined with temperature between 11 and 30 °C and relative humidity 53%–80%, where PR=1 exceeds 70%. 2) Increasing frequency of cleaning (both washing cleaning and vacuum cleaning) is effective for reducing risk level of mite infestation, but in humid environments (where relative humidity exceeds 53%) the effectiveness of cleaning is limited.