Analyzing Behavior of Agricultural Commodities in Kalimati Tarkari Market using Machine Learning Technique and Prediction Strategies

Abstract

Seasonal price volatility of agricultural commodities
poses significant challenges to farmers and market stakeholders
in Nepal due to climate variability, supply disruptions, and
limited access to predictive market information. This study
analyzes long-term seasonal price behavior using daily wholesale
price data from Kalimati Tarkari Bazaar spanning 2013–2023.
Machine learning and time-series techniques including Facebook
Prophet, logistic regression, STL decomposition, and K-means
clustering are employed to examine seasonal patterns, price
volatility, and future price trends across six Nepali seasons.
The results reveal strong and consistent seasonal dependencies,
with most vegetables exhibiting peak prices during winter (He
manta) and lower prices during the monsoon (Barsha). Prophet
based forecasting demonstrates moderate predictive performance
with an average MAE of 32.95 and an average R² of 0.42,
effectively capturing trend and seasonal components for most
commodities. Volatility analysis identifies high-risk commodities
with substantial price dispersion, while clustering reveals distinct
market segments based on price levels and variability. The
findings highlight the importance of seasonal awareness and data
driven forecasting in improving production planning, market
participation, and policy formulation in Nepal’s agricultural
sector.