Chapter 9 - Optimized adaptive tree seed Kalman filter for a diabetes recommendation system—bilevel performance improvement strategy for healthcare applications

Abstract

Over the past 2decades, recommendation systems have made remarkable changes and achievements in the field of disease prevention and control. Building an accurate recommendation system for various diseases is important. The recommendation systems primarily rely on data science and conform to standard real-time measures. Suitable data, when provided to recommendation systems, yield an excellent outcomes. The results of a recommendation system will also vary by changing its adaptability to the problem and input used. In this chapter, the authors define an optimized adaptive Kalman filter technique for diabetic recommendation systems. Despite the Kalman filter being best known for its efficacy in forecasting time changes and event prediction, its dependency on static attributes makes it somewhat restricted to real-time dynamic problems. Generally, healthcare problems are dynamic. In particular, the insulin requirement of a diabetes patient is an absolutely dynamic problem, as it is usually determined by various biological parameters that vary greatly with time. It is extremely difficult to recommend any information for a solution using a Kalman filter that relies on the static attributes of a real-time problem. Thus, as an attempt to remove the drawbacks in the existing Kalman filter, the authors propose an adaptive ability mechanism to the existing Kalman filter, which makes it very suitable for dynamic problems such as the insulin recommendation for diabetes. The adaptability performance is further boosted with the usage of a bioinspired optimization algorithm called the tree seed algorithm. This inherits the best values of features among the existing available features and allows only the best features to iterate to the next phase of the process, thus improving the process optimization. The usage of optimization algorithms in selecting the features for building an efficient recommendation gives an added advantage that it will not be affected by local minima solutions. A bilevel performance improvement strategy, such as the adaptive Kalman filter and the use of the tree seed algorithm, makes the proposed method robust in the field of diabetes recommendations. The results extracted from the proposed method are compared with those from conventional methods. The comparative analysis shows that the proposed method outperforms the existing methods in all performance indicators. The proposed method not only provides higher value performance indicators but also eliminates the tradeoff between higher performance indicators and the time taken.