Published Paper Details:

Multilayered feed forward neural networks have a variety of characteristics that make them particularly well-suited to challenges with complicated patterns. There is currently no training algorithm that can quickly and accurately determine an almost globally optimal set of weights, hence their use in real-world issues has been limited so far. As a class of optimization methods, genetic algorithms excel at intelligently exploring huge and complex spaces in search of values that are as close as possible to the global optimal. It is used in this paper to find the best weights for artificial neural networks (ANNs) using a genetic algorithm (GA). 78 questions were asked of 228 patients with first-time low-trauma hip fractures and 215 patients without a hip fracture in this study. We used logistic regression to select five relevant characteristics for building artificial neural networks that can predict the likelihood of a hip fracture (namely, bone mineral density, fracture experience, average hand grip strength, coffee intake, and peak expiratory flow rate).