Machine Learning–Enabled Prediction of Drug Release Kinetics in Controlled Delivery Platforms

Anber Abraheem Shlash Mohammad*¹, Suleiman Ibrahim Mohammad^2,3, Asokan Vasudevan^4,5, Sultan Alaswad Alenazi⁶, Badrea Al Oraini⁷

*¹Digital Marketing Department, Faculty of Administrative and Financial Sciences, University of Petra, Jordan mohammad197119@yahoo.com, ORCID: 0000-0003-3513-3965 ²Electronic Marketing and social media, Economic and Administrative Sciences Zarqa University, Jordan. ³Research follower, INTI International University, 71800 Negeri Sembilan, Malaysia. dr_sliman@yahoo.com, ORCID: 0000-0001-6156-9063 ⁴Faculty of Business and Communications, INTI International University, 71800 Negeri Sembilan, Malaysia. ⁵Shinawatra University, 99 Moo 10, Bangtoey, Samkhok, Pathum Thani 12160 Thailand asokan.vasudevan@newinti.edu.my, ORCID: 0000-0002-9866-4045 ⁶Marketing Department, College of Business, King Saud University, Riyadh 11362, Saudi Arabia. sualenazi@ksu.edu.sa ⁷Department of Business Administration. Collage of Business and Economics, Qassim University, Qassim – Saudi Arabia barieny@qu.edu.sa

Received: 14^th Aug, 2025; Revised: 15^th Sep 2025; Accepted: 15^th Nov, 2025; Available Online: 30^th Nov, 2025

ABSTRACT

Predicting drug release kinetics through controlled delivery platforms using machine learning helps overcome critical bottlenecks in pharmaceutical formulation design where Higuchi and Korsmeyer-Peppas non-linear models of hydrolysis-diffusion do not predict a non-linear interaction between hydrolysis-diffusion in electrospun nanofibers, PLGA microparticles, tablets and coaxial scaffolds. This paper incorporates supervised algorithms to simulate the role of fractional release F(t) of expansive in vitro measurements, focusing on tunable acetalated dextran (Ace-DEX) systems with cyclic acetal coverage (per cent of coverage to cost, CAC=20-80) and fibre diameter (200-100 nm). The workflow is anchored by the Gaussian Process Regression with isotropic Matern 5/2 kernels (sigmaL=0.734), trained on 929 observations using 10-fold cross-validation, augmented with SHAP interpretability to rank features (time IS=177.63), hybrid DTR-PAR-QPR to couple erosion, RF ensembles (n=500) to cluster Weibull, and PINNs (losses on PDE MSE pde=10 -4) to couple IVIV Hyperparameter optimization, LOO validation make parsimony, which prunes molecular descriptors. GPR has shown R²= 0.93 1 (RMSE= 0.084) beating empirics (p< 0.0001 ), and SHAP-directed generalisations cut assays 70-80% across payloads (paclitaxel to proteins). Hybrids give R 2 = 0.99887 to PLGA bursts, RF autocompletes T80 predictions (MAE=0.001), and PINNs provide in vivo gaps (R 2 = 0.92). These drug-agnostic platforms speed up QbD 10x and are the future of personalised therapeutics despite the lack of data ..

Keywords: Drug release, Machine learning, Gaussian Process Regression (GPR), SHAP analysis, Electrospun nanofibers, PLGA systems, Feature importance, IVIVC

How to cite this article: Mohammad AAS, Mohammad SI, Vasudevan A, Alenazi SA, Al Oraini B.; Machine Learning–Enabled Prediction of Drug Release Kinetics in Controlled Delivery Platforms. International Journal of Drug Delivery Technologies. 2025;15(4): 1853-1858, DOI: 10.25258/ijddt.15.4.38