📄 agriculture

Comparative Analysis of Exponential Smoothing and ARIMA Models for Forecasting Major Agricultural Crops in Bangladesh

This study evaluates Simple Exponential Smoothing, Holt's Linear Method, and ARIMA models to forecast production trends for rice, maize, potato, and wheat in Bangladesh, finding that Holt's Linear Method is optimal for three crops while ARIMA(0,3,2) best predicts wheat, thereby offering critical insights for future agricultural planning and policy-making.

Original authors: MAHIM MD. MUNTASHIR PRAM, Mohammad Shahed Masud

Published 2026-07-23

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: MAHIM MD. MUNTASHIR PRAM, Mohammad Shahed Masud

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to guess how many apples will fall from a tree next year. You could just guess, or you could look at how many fell last year, the year before, and the year before that. This is the heart of time series forecasting: using a history of numbers to predict the future. In the world of science, this is like being a detective for data, looking for hidden patterns in a trail of clues. Sometimes the trail is a straight line going up (a trend), sometimes it wiggles up and down like a heartbeat (seasonality), and sometimes it's just a messy scribble. To make sense of these scribbles, scientists use mathematical "crystal balls." Two of the most popular ones are Exponential Smoothing, which is like giving more weight to the most recent news while forgetting the distant past, and ARIMA, a more complex machine that tries to understand the deep, rhythmic connections between today's numbers and yesterday's. Why does this matter? Because if you are a farmer or a government leader, knowing whether you'll have enough food to feed a nation next year isn't just a math problem; it's the difference between a full belly and an empty one.

Now, let's zoom in on a specific detective story happening in Bangladesh. Two researchers, Mahim Md. Muntashir Pramanik and Dr. Mohammad Shahed Masud, decided to put three different forecasting tools to the test to see which one is the best at predicting the harvest of four major crops: rice, maize, potato, and wheat. They didn't just guess; they took a long look at historical data—some going back as far as 1961 for rice and as recently as 2000 for maize—and split it into two piles. One pile was the "training set," used to teach the models how to learn, and the other was the "validation set," used like a final exam to see how well the models actually performed.

The researchers tested three specific methods. First, there was Simple Exponential Smoothing (SES), a straightforward approach that smooths out the bumps in the data. Second, there was Holt's Linear Method, which is like SES but with a superpower: it can also spot if the numbers are generally going up or down over time (a trend). Finally, there was ARIMA, the sophisticated model that tries to capture complex patterns by looking at how past errors and past values influence the future. To decide the winner, they didn't just look at who guessed closest; they used a strict scoreboard of four metrics: Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Theil's U statistic. Think of these as different ways to measure how far off a guess was, with lower scores meaning a better prediction.

The results of this agricultural showdown were clear, though they weren't the same for every crop. For three of the four crops—rice, maize, and potato—the Holt's Linear Method emerged as the champion. It turned out that for these crops, the best strategy was to look at the recent history and the general upward trend, smoothing out the noise without overcomplicating things. The researchers found that by tuning the "smoothing parameters" just right, Holt's method consistently produced the lowest error rates. However, wheat was the rebel of the group. For wheat, the simpler methods couldn't quite catch the pattern. Instead, the ARIMA(0,3,2) model took the crown. This specific version of ARIMA, which involves a particular way of adjusting the data to make it stable, was the only one that could accurately forecast wheat production.

Based on these findings, the authors generated a ten-year forecast, looking ahead from 2022 to 2031. They suggest that for policymakers and farmers in Bangladesh, the key takeaway isn't that one single model is perfect for everything. Instead, the "best" tool depends entirely on the crop you are growing. If you are planning for rice, maize, or potatoes, the trend-focused Holt's method is likely your best friend. But if you are dealing with wheat, you need the more complex ARIMA machinery. The study doesn't claim to have solved the mystery of the future forever, but it does offer a reliable map for the next decade, helping stakeholders make smarter decisions to keep food secure and resources well-planned.

Technical Summary: Comparative Analysis of Exponential Smoothing and ARIMA Models for Forecasting Major Agricultural Crops in Bangladesh

Problem Statement
Accurate forecasting of agricultural production is fundamental to ensuring food security, optimizing resource allocation, and informing policy decisions in Bangladesh, where agriculture is a cornerstone of economic stability and employment. While various statistical techniques exist, there is a noted lack of comprehensive research systematically comparing the performance of different time series models across multiple staple crops within the specific context of Bangladesh. Existing literature often applies single methods (either ARIMA or exponential smoothing) to specific crops without a unified comparative framework. This study addresses this gap by evaluating and contrasting the forecasting capabilities of Simple Exponential Smoothing (SES), Holt's Linear Method, and Autoregressive Integrated Moving Average (ARIMA) models for four major crops: rice, maize, potato, and wheat.

Methodology
The study utilizes historical annual production data sourced from the Food and Agricultural Organization (FAO) of the United Nations. The datasets cover varying periods: rice (1961–2021), maize (2000–2021), potato (1990–2021), and wheat (1980–2021). The data were preprocessed to handle missing values and outliers, then partitioned into training and validation sets to ensure robust model evaluation.

Three distinct forecasting approaches were implemented:

Simple Exponential Smoothing (SES): A weighted average method utilizing a smoothing parameter ( $\alpha$ ) to emphasize recent observations.
Holt's Linear Method: An extension of SES that incorporates a trend component, utilizing two smoothing parameters ( $\alpha$ for the level and $\beta$ for the slope).
ARIMA: A model integrating autoregressive, differencing, and moving average components. The process involved checking for stationarity using the Augmented Dickey-Fuller (ADF) test, identifying orders ( $p, d, q$ ) via ACF and PACF plots, and selecting the optimal model based on the minimum Akaike Information Criterion (AIC).

Model selection and performance assessment were conducted using a grid search method to optimize smoothing parameters for SES and Holt's methods. The forecasting accuracy of all models was rigorously evaluated using four statistical metrics: Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Theil's U statistic. Diagnostic checks, including residual analysis and normality tests (QQ-plots), were performed to validate model assumptions.

Key Results
The comparative analysis yielded distinct optimal models for different crops based on the performance metrics:

Rice, Maize, and Potato: Holt's Linear Method demonstrated superior performance over both SES and ARIMA for these three crops. The results indicate that incorporating a trend component significantly enhances predictive accuracy for these specific agricultural time series. Optimal smoothing parameters were identified for each crop to minimize error.
Wheat: The ARIMA model, specifically the ARIMA(0,3,2) configuration, outperformed the exponential smoothing methods for wheat production forecasting.

Following model selection, the study generated ten-year forecasts (2022–2031) for each crop using the best-performing models. The diagnostic plots for the selected models (e.g., ARIMA(1,1,1) for rice) confirmed that residuals were randomly distributed, normally distributed, and free of significant autocorrelation, validating the reliability of the fitted models.

Significance and Claims
The paper asserts that the selection of an appropriate forecasting model is critical and must be tailored to the specific characteristics of the crop data. By identifying that Holt's Linear Method is generally more effective for rice, maize, and potato, while ARIMA is superior for wheat, the study provides a data-driven basis for model selection in agricultural planning.

The authors claim that these findings offer valuable insights for researchers, policymakers, and agricultural stakeholders in Bangladesh. The primary significance lies in the potential to enhance decision-making and agricultural planning through more accurate production forecasts, thereby supporting strategic resource management and food security initiatives. The study does not propose new theoretical frameworks but rather emphasizes the practical application of established statistical methods to improve the precision of agricultural forecasting in the region.

More like this