ANFIS and LSTM Techniques based Model for Time-Series Forecasting

One of the most challenging issues in the world of data science and machine learning is time series forecasting. A set of continuous variables must be predicted over a range of time intervals. In order to fit the curve to the time series and generate forecasts, traditional methodologies call for the analytical computation of regression factors. Such techniques take a lot of time, are prone to mistakes, and do not accurately represent the facts. Time series forecasting has gained a lot of interest recently because of advancements in the disciplines of deep learning (DL) and artificial intelligence (AI), particularly in the DL space. This research expands on prior work on time series forecasting that used pricing from petrochemical goods as a case study and explored the deep learning approach as a soft computing model vs ANFIS (Adaptive Neuro-Fuzzy Inference System) as a classical model for anticipating resin prices. This research suggests utilizing a hybrid neuro-fuzzy model with Long Short-Term Memory (LSTM) and ANFIS to perform forecasting as well as compare their results. Predicting outcome demonstrates that the LSTM model has a relatively low error in terms of SMAE that is 30.48 for night time data, 6.27 for unemployment data, and 110.26 for agriculture, and also a variance score which is 0.92 for night time data, 0.55 for unemployment data, and 0.63 for agriculture of India while SMAPE that is 30.96 for night time data, 3.81 for unemployment data, and 57.05 for agriculture and also a variance score which is 0.95 for night time data, 0.84 for unemployment data, and 0.79 for agriculture of Uttarakhand. Moreover, more performance metrics are used, and their test also demonstrates that there is an important difference in LSTM based estimating accuracy if contrasted to LSTM and ANFIS.