Improving the Accuracy of PCOS’ Data Prediction Model Based on Data Balancing and Multilayer Feature Selection Strategies

Abstract

The use of machine learning algorithms to design a model for PCOS prediction or diagnosis in a recent time is attracting an impressive magnitude of research attentions. Of course, the rationale behind this development is not farfetched as the disease is common in women of reproductive age, which causes infertility. Also the algorithms as top notch techniques for classification, are highly promising. However, attention has to be paid to the raw dataset used for training the models; this is because the strategies adopted for feature engineering processes have direct proportionate effect on the robustness state of the model. Meanwhile, significant efforts have been advanced towards this development but that does not foreclose adoption of more strategies for better accuracy of the model. To this end, this research aims to adopt multilayers strategies in terms of class balancing and feature selection to improve the accuracy of the existing PCOS’s data. The strategies include the use oversampling and LASSO-Pearson’s correlation techniques for class balancing and feature selection respectively for the proposed ensemble random classifier based model. The 97.80% accuracy result of the proposed model outperforms the rest of the other seven models used in the benchmark work. Therefore, careful attentions have to be constantly advanced towards the process of feature engineering, which include data preprocessing, data exploratory, data balancing and feature selection strategies for optimal result.