Agriculture Commodities Price Prediction using

Advanced Time Series Techniques

TEAM NUMBER: 6
NAME OF TEAM MEMBERS : NAME OF THE GUIDE:
A.C Aravind - 18H61A0504 Mr.D.RamanaKumar, M.Tech(CSE),(Ph.D.)
K. Anand Kumar - 19H65A0501 Department of Computer Science Engineering
A. Adithya -19H65A0505
 Contents
• Abstract
• Introduction
• Literature Survey
• Motivation
• Existing Approaches and Drawbacks
• Aims & Objectives
• Methodology
• Tools & Technologies
• Design
• UML Diagrams
• Implementation
• Conclusion
 Abstract
Price prediction of agricultural commodities is the essential requirement for the
sustainable development of the farming community. This forecasting methodology is
very important for sustainability of farmers. The accurate forecasting of agricultural
prices will help in taking better decisions which results in minimizing the loss and
managing the risk of price fluctuations. This methodology provides valuable insights
that can help the farmers to reduce cost, restrict the use of chemical fertilizers,
increase revenue, understand the seasonal trend demand, etc., We focus on the short-
term and long-term forecasting of agriculture commodities prices. We will develop
independent advanced forecasting models using time-series analysis like SARIMA,
DeepAR for accurate price prediction of agricultural commodities.
 Literature Survey
Gan-qiong Li a,b*, Shi-wei Xu a,b, Zhe-min[1], time series analysis has been often employed in
modeling short-term price forecasts. In recent years a new technique of artificial neural networks ANN
has been proposed as an efficient tool for modeling and forecasting. A feed-forward ANN model has
been developed for short-term price forecasting of tomato and in comparison with time series model
ARIMA in this study.

Girish K. Jha and Kanchan Sinhab[2], In this study, the superiority of ANN over linear model
methodology has been demonstrated using monthly wholesale price series of soybean and rapeseed-
mustard. The empirical analysis has indicated that ANN models are able to capture a significant number
of directions of monthly price change as compared to linear models.
Mr. R. Parasuram, Mr. Mangesh Tyagi[3], used machine learning models to predict (short-term) prices of
Soyabean. The algorithm was created to predict modal prices at 14-day intervals across the major
mandis in MP. It achieved an accuracy rate of 95%.
 Introduction
• Agriculture and cultivation have been part of human civilization for centuries and
is evolving with technology, giving rise to smart farms.
• In the present era of globalization, management of food security in the agriculture
dominated developing countries like India needs efficient and reliable food price
forecasting models more.
• Currently, India is ranked second in the world for production of agricultural
commodities and contributes almost 18% in the Indian GDP.
• Based on a forecast, the insurance companies and government can support the
farmers financially by providing insurance policies and loans at affordable interest
rates.
• The prices of crops are volatile, especially for tree crops and the crops cultivated
annually. Unpredictable events such as drought and flood can affect the prices of
agricultural commodities, thus affecting the entire market; the farmers, suppliers,
exporters, and stakeholders face huge losses.
• This proposed model can predict the price of agricultural commodities on the
following hour, day and also predict if the price on the next day will
increase/decrease or stay stable in near future.
 AIMS AND OBJECTIVES

• To investigate Agricultural commodities price forecasting based on

machine learning methods (SARIMA, DeepAR)
• Short-term forecasting of agricultural commodities prices, covering the 1-
hour, 1-day ahead and 1-week ahead forecasts and long-term forecasting
covering monthly and yearly forecasts.
• To predict the prices of agricultural commodities on the following hour,
day and also predict if the price on the next day will increase/decrease or
stay stable in near future.
 Motivation

• To provide accurate and timely price forecast by taking into account (Local
information to the farmers, traders and policy makers) so that they can
make production, marketing and policy decisions well in advance.
• The ability to predict crops before the start of the crop season.
• It'll be a risk for farmers to grow any crop without foreseeing the demand,
price etc.
Existing Approaches
• Due to significant economic effects of agricultural commodities price forecasting,
many techniques have been investigated, statistical approaches like support vector
machines(SVM) and ARMA (autoregressive moving average).By using Time Delay
Neural Network(TDNN), a special form of ANN; provided inconsistent results
while dealing with seasonal datasets.

Drawbacks
• In existing system, most of the researchers dealt with agricultural price forecasts but
they relied on monthly forward products mostly instead focusing on spot market
prices. However, publications in the domain of agricultural forecasting is scarce.
 Methodology

Data Pre-processing
Data Loading
• Handling outliers
• Commodities data
• Handling Missing Values
• Weather data
• Dropping unwanted columns

Feature Extraction
• Selecting the features and
Data Splitting

Building the model

• Random Forest
Regressor

User input
Time Series Analysis Visualization • Commodity name
• SARIMA Model • Matplotlib
• Date (Long Term)
• DeepAR Algorithm • Seaborn • Date and Hour (Short Term)

Displaying
Predicted Result
 Tools & Technologies

MINIMUM SOFTWARE AND HARDWARE REQUIREMENTS: 

• Anaconda 3.7
• Operating system : windows, Linux
• Processor : intel i3
• Ram : 4gb
• Hard disk : 250gb
• Frontend Framework : Flask
DESIGN
Architecture
 Workflow

Importing Loading Data Pre- Feature

Packages Datasets processing Extraction

Model Building
(Random forest
regressor)
SARIMA
Time series
Visualization Analysis

DeepAR
Choose
commodity and
date

Predicted
Result
Use case Diagram
State Chart Diagram
Activity Diagram
Implementation
 Module Description
System Module
• Data Pre-processing: Handling outliers and missing values.

• Model Building:
It uses forecasting such as Random Forest Regressor, SARIMA, DeepAR for accuracy
forecasting.

• Data Visualization:
System makes use of seaborn and matplotlib for visualization of trends, seasonality and displays
predicted result.

User Module
• Loading Datasets:
Commodities Dataset, Rainfall Dataset

• Choosing commodities:
Commodities such as paddy, cotton, mangoes, sugarcane, maize

• Choosing Date and Hour for short-term and long-term forecasts.

 Sample Code
Installing and importing dependencies

Loading the Data

 Sample Code

Converting Non Stationary Time Series into Stationary

Non Stationary Time Series

 Sample Code
Code

Output
 References

[1] George S. Atsalakis and Kimon P. Valavanis. Forecasting Stock Market Short-term Trends
Using a Neuro-fuzzy Based Methodology. Expert Systems with Applications, Volume 36, Issue
7, September 2009.
[2] Farzan Aminian, E.Dante Suarez, Mehran Aminian and Daniel T. Walz.
Forecasting Economic Data with Neural Networks. Computational Economics, 28:
71-88, 2006.
[3] Hagan, M. T. and Menhaj, M. (1994) Training feed-forward networks with the
Marquardt algorithm. IEEE Transactions on Neural Networks, 5: 989-993.
[4] McLeod, A.I. and Li, W.K. (1983) Diagnostic checking ARMA time series
models using squared residual autocorrelations. Journal of Time Series Analysis, 4:
269-273.
[5] Zhang, G., Patuwo, B. E. and Hu, M. Y. (1998) Forecasting with artificial neural
networks: The state of the art. International Journal of Forecasting, 14: 35-62.
 Conclusion

Forecasting agricultural commodities prices 1 hour,1 day ahead and 1 week

ahead for short-term forecasting and long-term forecasting covering monthly
and yearly forecasts which has adverse effects on farmers, stake holders .etc
Thank you!