Time-Series Data Prediction Using Johnson-Lindenstrauss Lemma, Fuzzy Logic, And Self Organizing Maps

This paper uses a hybrid model combining Self-Organizing Maps (SOM), Johnson-Lindenstrauss Lemma(JLL), and Fuzzy Logic for time-series data prediction. It is named as SJLF model. SOM is used to group data having similar characteristics. By JLL, high-dimensional data is projected to low-dimensional space by approximately preserving the distance between the input vectors. The fuzziness in data is also carried on to the projected values. These projected values are input into the fuzzy logic system to obtain the predicted value as output. For experimental analysis, the available data on COVID-19 is taken from humanitarian data exchange. The proposed SJLF model is applied to five coronavirus-affected countries Belgium, Brazil, Colombia, India, and Iran. The SJLF model's prediction shows promising results as the average MAPE for five countries is 1.199, and the prediction accuracy on an average is 98.8%. The proposed model is compared with the ANFIS model and is found that the proposed model shows better forecasting results.