V3I7P64

Neural-Network-Based Mathematical Model for Optimizing the Success of Public–Private Partnership Irrigation Projects: Evidence from the Galana–Kulalu Scheme, Kenya 

Patrick Mugendi Njagi^1*, Patrick Ajwang¹, Charles K. Kabubo¹

Abstract

Public–private partnerships (PPPs) are increasingly relied upon to finance large irrigation schemes in developing economies, yet many such projects underperform on cost, time and scope despite the presence of well-documented critical success factors (CSFs). A recurring weakness of the empirical PPP literature is its reliance on linear statistical tools that struggle to capture the nonlinear, interacting and often compensatory ways in which CSFs shape project outcomes. This study formulates and trains a feed-forward artificial neural network (ANN) as a mathematical model that maps the CSFs of a large irrigation PPP onto a composite success index and then uses the trained surface to identify success-maximizing intervention profiles. The model architecture follows the conceptual framework of a survey study of the Galana–Kulalu irrigation scheme in Kenya, in which planning-stage CSFs, implementation-stage CSFs and general (legal, risk-allocation and governance) CSFs act as inputs, the institutional framework enters as a moderating input, and success is measured through cost, time and scope. A calibrated dataset of 4,000 observations was used via by Monte-Carlo sampling. A 26–16–1 network trained with the Adam optimizer explained 87.8% of the variance in the hold-out set (R² = 0.878, RMSE = 0.337, MAPE = 8.3%), outperforming a multiple-linear-regression benchmark (R² = 0.845). Connection-weight analysis attributed 36.0% of predictive importance to general CSFs, 28.0% to implementation CSFs, 24.7% to planning CSFs and 11.3% to the institutional framework. A targeted scenario in which the three most influential challenges were mitigated by one Likert point and the three leading enablers strengthened by one point raised the predicted success index from 3.55 to 4.22, an improvement of 18.8%. The model provides irrigation-sector decision-makers with a transparent, retrainable optimization tool, and its Python and MATLAB implementations are reported in full.

Keywords:

Public–private partnership; Artificial neural network; Critical success factors; Irrigation projects; Success optimization; Galana–Kulalu