Thesis (MSc) - Cyprus International University. Institute of Graduate Studies and Research Computer Engineering Department

Includes bibliography (sheets 50-54)

ABSTRACT
Malaria is a very serious and sometimes a deadly sickness resulting from a parasite
that is commonly found in female anopheles' mosquitos which feed on human beings.
Humans that contract malaria are usually very sick with flu-like contamination, very
high fevers, and shaking chills. The highest rates of malaria infections are found in
African countries. This research was done to provide a cost effective machine learning
method to perform malaria diagnosis in Africa especially in the rural areas. The
research of this study was carried out using the following methodology. 27,558 images
containing equals parts of infected and uninfected blood cells then were further divided
into test, training and validation sets. The model was created using convolutional
neural networks(CNN) which part of deep learning. Five layers of convolution were
used with max pooling layers between the convolution layers. Binary crossentropy was
used to calculate the loses. To evaluate the custom CNN model, the metrics measured
were accuracy, f1 score, precision and recall. The results of the research showed that
the proposed custom CNN model had 96.1% accuracy, an F1 score 0.96, precision
score 0.968 and a recall score of 0.95. These results indicated a successful custom
CNN implementation. The conclusions and future recommendations are also discussed
in this research to further improve accuracy and help the fight against malaria in the
rural areas of Africa.
Keywords: CNN, Deep learning, Machine learning, Malaria