A TEK-Integrated Decision-Support Framework for Traditional Lift-Net Fisheries: System Design and Simulation-Based Feasibility Analysis

Abstract

Deploying machine-learning decision support in traditional fishing is constrained less by predictive accuracy than by whether practitioners accept and appropriately use advice alongside their own tacit knowledge. This paper presents the design of a decision-support framework for traditional Chinese fishing-net (Cheena vala) operations in Kerala, India, that couples a Random Forest catch-suitability model with formalized Traditional Ecological Knowledge (TEK) while preserving fisher decision authority, together with a simulation-based feasibility analysis. The prediction component is evaluated on a 1,000-event environmental dataset emulating the study site: a Random Forest model attains a mean absolute error of 0.84 kg (R² = 0.22) for catch-weight prediction, outperforming Gradient Boosting, an LSTM, and a hybrid TEK-ML variant, with sonar fish-detection the dominant feature (34.7% importance). Fisher-adoption behaviour is examined in a transparent simulation testbed driven by an explicit stochastic behaviour model with stated parameters; it is intended to surface framework properties and to specify hypotheses for subsequent field validation, and does not constitute empirical evidence about real fisher behaviour. Under the stated model, overall simulated adherence is 74.9%, a smart-waiting policy reduces modelled operating cost through avoided low-suitability trips, and the hybrid model improves catch-weight error by 7-14% where TEK rules are active. We articulate design principles for decision support in traditional contexts and a proposed technology-acceptance model framed as testable hypotheses.