Statistical Evaluation on the Performance of Dyslexia Risk Screening System Based Fuzzy Logic and WEKA

Abstract

This paper presents a comparative study based on statistical analysis of dyslexia risk screening classification implemented using Fuzzy Inference System (FIS) and Waikato Environment for Knowledge Analysis (WEKA) data mining tool. A rapid dyslexia screening system based FIS has been previously developed using four dyslexia screening tests with initial objective to differentiate between normal, dyslexia and slow learner. However, the slow learner subjects were removed from the collected data because of its similar tests scores with dyslexia subjects. Using the collected data (n=104), the dyslexia risk screening via FIS is able to classify normal and dyslexia subjects. A comparative study based statistical evaluation is performed using Naïve Bayes, Decision Table and Random Forest (WEKA) to compare their accuracy, sensitivity, specificity and precision respectively. FIS has achieved 100 % for all the statistical performance in classifying dyslexia subject showing great improvement compared to the previous developed system. Meanwhile, all the three classifiers (WEKA) produce 100 % in terms of specificity and precision using training datasets. It is observed that the Random Forest (with training dataset n=200) shows the best classification performance among three classifiers in WEKA as it managed to reach 100 % in accuracy and sensitivity performance. This paper concludes that FIS and Random Forest are suitable to be used for dyslexic risk confirmation as both yield the best statistical performance. Future studies cover the development of standalone application and evaluation on the user acceptance and satisfaction along with the addition of IQ test in the second stage of FIS.