Inflation Prediction of Agricultural Products Using Fuzzy Inference System.
Dataset: Wholesale price for agricultural commodities (WPI).
Objective: Devise time series prediction techinques which would work for small data set.
Research Paper published at 11ICCCNT is included in the repository.
An Android Application named PREDICTR is also under development.
Abstract Predictions based on Sequential Data such as timeseries data of agricultural product prices play a crucial role in agriculture-based business. Determination of inflation in prices help farmers and associated businesses to take corrective measures for higher returns. However, unavailability of enough collective and accurate data for Indian Markets challenges accuracy. This paper captures the advantage of NN (Neural Networks) and FZ (Fuzzy Systems) for predictions based on time series analysis with limited data. NN learn by adjusting the weights between connecting neurons. This helps in pattern recognition of similar data points. Recent developments in DL (Deep Learning) such as the RNN (Recurrent Neural Network) variant, LSTM (Long Short Term Memory) dominates the trade market predictions. LSTM solves the gradient descent problem of traditional NN and remembers temporal patterns. FZ, on the other hand, helps in making inference about human cognition through membership functions. Learning capabilities of NN and Fuzzy rules form the novel Neuro-Fuzzy system termed as FLSTM (Fuzzy-LSTM). Further, the data set contains monthly wholesale prices published by the Ministry of Commerce and Industry, Govt. of India for essential agricultural commodities. The evaluation based on the proposed work shows decent improvement than some standard DL model for various entities when subject to limited records.
Supporting images
Comparison of LSTM and Fuzzy Models for Onion data set.
Results over Onion prices dataset.
Work Flow of the system(with working prototype of Android Application).
Android Application Sample Image(Generating prediction graph for Onion Data set).
