Performance analysis of Machine learning classifiers for predicting diabetes and prostate cancer

Healthcare industry have always been a big resource of medical data. The size of databases increases rapidly every day but still not well explored to discover hidden knowledge. Advanced machine learning techniques can be used to develop predictive models for medical data, such models could be very useful for physician to make the decision. This paper discusses the comparison of some popular machine learning algorithms in the aim to find the best predictive model for medically related problems that we can use to develop our future medical ”workflow”. The case study addressed in this work involves predicting diabetes and prostate cancer using eight different machine learning architectures in order to investigate their accuracy and performance. The results obtained during our experiments have shown that of the studied models, Recurrent Neural Network produced higher results in comparison with other models, but was closely followed by Logistic regression which produced very good results especially with scaled features. To evaluate the performance of each model, we use three measures which are accuracy, AUC and F-measure in order to have a more complete evaluation and avoid cases where the use of only one metric could be misleading. According to the three measure results, RNN have produced the most equilibrated results.