Building Machine Learning Algoritms

We are a group of doctors, engineers and researchers working on a project to combine machine learning and clinical medicine to improve clinician judgement. This group is affiliated with two of the most well known univerisites in the Nordic. The aim is to to better understand AI, ML + medicine combo in order to improve diagnosis or calculate probabilities/risk of different diseases and outcomes. We have access to datasets from real patients and will be working on these. For these projects we would need a programmer/engineer who has a good understanding of mathematics and machine learning. On the datasets we have we are looking to utilize and compare different algorithms, both supervised and unsupervised. Furthermore, we will be researching how to increase explainability by using Shapley Additive Explanaitions (SHAP) and other techniques such as Local Interpretable Model Agnostic Explanations etc. An example: We have a small but high quality dataset of 105 with patients with middle ear infection who were followed up during 3 years. Different parameters (such as age, heredity, siblings, parental smoking status etc) and bacterial cultures (date of culture, type of bacteria, type of antibiotics if any) were collected. These will be the input values. Output will be either binary or regression. Here, we are interested in comparing different algorithms (AUC, accuracy etc.) as well as SHAP values (summary-, force- and dependency plots). In this particular case the aim is not to create an algorithm for accurate prediction, it is rather to compare and highlight the potential differences of algorithms in real life judgements on population- and individual level. We are looking to better understand the characteristics that are important when designing algorithms for clinical purposes, depending on the setting. We have access to bigger datasets for other diseases and look forward to working on these when we have found a suitable programmer.