A Comparative Study of Stand-Alone and Hybrid CNN Models for Malaria Parasite Detection

A substantial percentage of deaths from malaria, a deadly illness spread by mosquitoes, occur in tropical and subtropical areas as a result of poor detection methods, insufficient laboratory experience, and other issues. In areas with little resources, detecting malaria poses a serious problem. Here, a unique method that uses machine learning (ML) algorithms for prediction and deep learning (DL) models for feature extraction has been explored. In particular, the use of random forest (RF), support vector machine (SVM), and k-nearest neighbours (KNN) classifiers in conjunction with convolutional neural networks (CNN) for feature extraction as well as prediction has been investigated. Custom CNN architectures were devised for feature extraction that was subsequently fed into ML classifiers for prediction. This approach aims to enhance malaria parasite detection accuracy, particularly in resource-limited settings, by leveraging the strengths of DL for feature extraction and ML for prediction. The findings demonstrate that using hybrid models instead of the standalone CNN model does not significantly improve performance. With 95.45% model accuracy, the CNN+SVM models perform the best. All four of the models that are being given are robust, with 95.05% being the lowest accuracy achieved.