Smart Agricultural Computing Development of Robust Data Processing and Machine Learning Pipeline for Durian Price Prediction

This study presents a machine learning framework for predicting durian prices using IoT environmental sensor data. The research integrates six key environmental variables such as temperature, humidity, solar, wind speed, wind direction, and rainfall with daily Grade AB durian export prices collected during the peak harvesting seasons of 2023-2024. The study implements three preprocessing approaches consisting of normal processing, 120-day Historical Average Window (AVG120) and 90-day Shift Moving Average with 30-day Forward Prediction (SMA90). Using Random Forest as the baseline model, the results demonstrate that moving average methods significantly outperform traditional preprocessing. The SMA90 method achieved the highest accuracy with R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> = 0.9778 and lowest error rates (MSE = 0.0015, RMSE = 0.0342) in cross-validation while AVG120 showed comparable performance (R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> = 0.9721). This framework provides farmers with a valuable tool for data-driven decision-making in agricultural technology planning and market timing.