Predictive self-organizing neural networks for in-home detection of Mild Cognitive Impairment

Abstract

In-home sensing of daily living patterns from older adults coupled with machine learning is a promising approach to detect Mild Cognitive Impairment (MCI), a potentially reversible condition with early detection and appropriate intervention. However, the number of subjects involved in such real-world studies is typically limited, posing the so-called small data problem to most predictive models which rely on a sizable number of labeled data. In this work, a predictive self-organizing neural network known as fuzzy Adaptive Resonance Associate Map (fuzzy ARAM) is proposed to detect MCI using in-home sensor data collected from a unique Singapore cross-sectional study. Specifically, mean and standard deviation of nine in-home behavioral attributes of 49 subjects over two months were derived for each subject from the raw sensor data. We first applied fuzzy ARAM to the 49-subject data set with missing data, and achieved a F1-score of 58.3% to detect MCI from cognitive healthy. To eliminate the effect of missing data, we next conducted our study using an even smaller 25-subject data set with no missing values, of which fuzzy ARAM achieved a F1-score of 63.6%. To derive concise rules for prediction and interpretation, antecedent pruning was subsequently employed. For the 25-subject data set, the F1-score improved to 76.2%, while the symbolic IF–THEN rules revealed that behavior metrics such as variation of forgetfulness and sleep contained notable predictive utility. Compared with Support Vector Machines (SVM), Decision Tree, Multi-Layer Perceptron (MLP), Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM), our benchmark experiments show that fuzzy ARAM provided the highest predictive performance and yielded unique rules for MCI detection. These results demonstrate the potential of fuzzy ARAM to detect MCI using in-home monitoring sensor data.