Clustering and Classification of a Qualitative Colorimetric Test

In this paper, we present machine learning based detection methods for a qualitative colorimetric test. Such an automatic system on mobile platform can emancipate the test result from the color perception of individuals and its subjectivity of interpretation, which can help millions of populations to access colorimetric test results for healthcare, allergen detection, forensic analysis, environmental monitoring and agricultural decision on point-of-care platforms. The case of plasmonic enzyme-linked immunosorbent assay (ELISA) based tuberculosis disease is utilized as a model experiment. Both supervised and unsupervised machine learning techniques are employed for the binary classification based on color moments. Using 10-fold cross validation, the ensemble bagged tree and k-nearest neighbors algorithm achieved 96.1% and 97.6% accuracy, respectively. The use of multi-layer perceptron with Bayesian regularization backpropagation provided 99.2% accuracy. Such high accuracy system can be trained off-line and deployed to mobile devices to produce an automatic colourimetric diagnostic decision anytime anywhere.