A coherent understanding of plantar pressures is crucial in the field of sports research and clinical studies. Real-time and dynamic analysis of plantar pressures require sensors that are in a wearable form factor. Sensor insoles built on a variety of principles satisfy this condition. The accuracy of total force provided by commercially available sensor insoles is significantly low compared to that provided by force plates, the gold standard for force measurements. The current work discusses the limitations of insole based force measurement from the design standpoint alone. We further propose a learning-based model to improve the accuracy of the capacitive sensor based insoles for the task of total force measurement. A learning-based model is developed by training on data collected from sensor insoles for loads varying from 0 to 80 kg on each insole. The absolute mean percentage error of left and right soles of specific sized capacitive insoles as compared to the gold standard has reduced from 8.01 to 5.19 and 35.85 to 21.64 respectively. The learning based model has shown promising results providing a reliable scope towards the extension of the work under variable foot anatomies and pressure distributions. © 2019 IEEE.