Predictive Modeling of Diabetic Kidney Disease using Random Forest Algorithm along with Features Selection

Artificial Intelligence for the Diagnosis and Treatment of Diabetes Kidney Disease: a systematic review

“…If the AI model even outperformed the conventional statistical method the cell is filled with a 'yes'. Except for two studies (59,45), all 22 other included studies either outperformed conventional methods or accomplished the result which is not attainable through non-AI frameworks. This proves that AI algorithms are superior to conventional methods, and they might be suitable in clinical practices to aid physicians.…”

“…The details of the selected articles are presented in Table -3 The scheme of this study is summarized into four different categories based on the AI framework in use. The categories include 1) predictive AI model for the early detection of DN (40,44,45,46,48,51,54,55,59,61), 2) AI Framework for the diagnosis of DN (42,50,57), 3) predictive model for the progression of DKD (39,41,49,58,62), 4) the management of DN for existing patients OR early management indication to DM patients without DKD with the help of AI (43,47,52,53,56,60). The studies included in this systematic review are from various parts of the world.…”

“…A total of 8 studies (39,40,44,45,48,49,53,59) In the quality assessment (Table 4) the shortcomings OR strengths of the studies were assessed. To ensure the included study adds something new to clinical practices and fulfils a void that cannot be assessed by conventional statistical methods, the "limits in current non-AI approach" column is present in Table-4.…”

“…Meanwhile, 7 studies(41,50,51,52,54,55,57) included a cohort size/ sample size over 10,000 patients/sample size. In addition, a large sum of 14 studies(39,40,42,44,48,49,51,52,53,55,56,57,59,62) used ML to build the predictive model. On the other hand, 5 studies(47,50,54,58,60) used DL algorithms as the AI method in their study.…”

“…It is worth noting that a large number of the studies used EHR datasets from a regional hospital/clinic. A total of nine studies(42,43,46,51,52,54,58,59,61) used an EHR dataset from a regional hospital.As most of the included studies are predictive models, they require a certain criterion to be met. The data should be pre-processed with the necessary imputation, dimensionality reduction, transformation etc.…”