Of all cancers occurring in women, breast cancer is the most common. 2 million new cases and over 0.5 million deaths globally each year. In the UK alone, between 2015 and 2017, 11,400 deaths each year, or 31 per day. Globally, many doctors are overwhelmed, having to deal with up to 70 patients a day. Mistakes in diagnosis have deadly consequences. These are shocking, extremely sad statistics, especially as all 100% of women with the disease can survive their first year if diagnosed at the earliest stage.

Disease diagnosis implies complex data involving several levels of uncertainty and imprecision. A single disease can manifest itself differently, with varying intensities, depending on the patient. A single symptom may correspond to different diseases. On the other hand, several diseases present in a patient may interact and interfere with the usual description of any one disease.

The report and source code linked below presents a Python-based fuzzy logic knowledge-based inference system for the binary classification of breast cancer tumours. Fuzzy logic lends itself to modelling domains where vagueness and blurriness are present. The solution takes an extensive data-driven, supervised learning approach in defining linguistic terms, implication rule sets and membership functions, with the Wisconsin Diagnostic Breast Cancer dataset used as the empirical, impartial problem domain expert.

The best FIS model classified all test set malignant tumors correctly with an overall accuracy of 93.6%, outperforming the logistical regression, decision tree, random forest machine and neural network learning methods. Fuzzy logic systems are transparent, simple and interpretable. They can provide a far more “explainable A.I.” solution than machine learning alternatives such as neural networks.

Master’s Degree Intelligent Systems paper grade 98% | Source code