No description available

Learning About Our Dataset Diabetes The increasing prevalence of diabetes in the 21st century is a problem. Patients have symptoms like unusual thirst frequent urination extreme fatigue Diabetes can also lead to more serious complications like stroke, blindness, loss of limbs, kidney failure, and even heart attack. Discovery of Insulin In the 1920s, insulin was discovered by Frederick Banting. Most of the food we eat is turned to glucose, or sugar, for our bodies to use for energy. The pancreas, an organ near the stomach, makes a hormone called insulin, to help glucose get into the cells of our bodies. When you have diabetes, your body either doesn't make enough insulin or can't use its own insulin as well as it should. And this causes sugars to build-up in the blood. With Banting's discovery of insulin, pharmaceutical companies began large-scale production of insulin. Although it doesn't cure diabetes, it's one of the biggest discoveries in medicine. When it came, it was like a miracle. Challenges with Insulin The default method of administration is by a needle, multiple times a day. Insulin pumps are a more recent invention. These are insulin delivering devices that are semi-permanently connected to a diabetic's body. The Future: Oral Insulin? Wouldn't it be great if diabetics could take insulin orally? This is an active area of research, but historically the roadblock is getting insulin through the stomach's thick lining. Our dataset: Auralin and Novodra Trials We will be looking at the phase two clinical trial data of 350 patients for a new innovative oral insulin called Auralin - a proprietary capsule that can solve this stomach lining problem. Phase two trials are intended to: Test the efficacy and the dose response of a drug Identify adverse reactions In this trial, half of the patients are being treated with Auralin, and the other 175 being treated with a popular injectable insulin called Novodra. By comparing key metrics between these two drugs, we can determine if Auralin is effective. Why do we need Data Cleaning? Healthcare data is notorious for its errors and disorganization, and its clinical trial data is no exception. For example, human errors during the patient registration process mean we can have duplicate data missing data inaccurate data You're going to take the first step in fixing these issues by assessing this data sets quality and tidiness, and then cleaning all of these issues using Python and Pandas. Our goal is to create a trustworthy analysis. DISCLAIMER: This Data Isn't "Real" The Auralin and Novodra are not real insulin products. This clinical trial data was fabricated for the sake of this course. When assessing this data, the issues that you'll detect (and later clean) are meant to simulate real-world data quality and tidiness issues. That said: This dataset was constructed with the consult of real doctors to ensure plausibility. This clinical trial data for an alternative insulin was inspired and closely mimics this real clinical trial for an inhaled insulin called Afrezza. The data quality issues in this dataset mimic real, common data quality issues in healthcare data. These issues impact the quality of care, patient registration, and revenue. The patients in this dataset were created using this fake name generator and do not include real names, addresses, phone numbers, emails, etc. The video above is only a short preview of the dataset that is intended to motivate. If you're not comfortable with the meanings of each column in each table, please revisit the Visual Assessment: Acquaint Yourself page in Lesson 3: Assessing Data. Descriptions of each column as well as the Auralin clinical trial, as a whole, are presented there.

The data set chosen from the UCI repository pertains to heart failure clinical records. The purpose of this report is to be able to make predictions about patient survivability by making use of the patients clinical records like pre-existing health conditions like diabetes, hypertension or anemia, patient’s habits like smoking and patient’s vital like serum creatinine, creatinine phosphokinase, serum sodium and other such indicators. For this purpose the data was prepared in task 1 using pandas library by transferring it to a data frame. In task 2 the data was explored using boxplot, histogram, pie charts and bar graphs. Relationships between features were explored and visualized and a meaningful question about patient survival rate depending on pre-existing health conditions was posed and answered using pie charts and group by method. In task 3 feature selection was undertaken and ‘time’ feature was eliminated because it can vary according to heart failure outcome i.e. be less for a patient when the patient died during follow-up. Two predictive models Decision Tree Classifier and K-Nearest Neighbor Classifier were chosen for the data set. Since KNN is a distance-based algorithm (utilizing Minkowski’s distance) X was scaled using the MinMaxScaler function to the default range of 0, 1 to get best possible results for KNN. While for decision tree the original values themselves were utilized. Before fitting the models the values of hyper parameters for both KNN (n_neighbors) and decision tree (max_depth, min_samples_leaf) were tuned. The classifiers were initialized using the tuned hyper parameters and then the predictions are obtained. To evaluate the performance of the predictive models the predictions were compared to the ground truth using accuracy scores, confusion matrices and performance report. These were then analyzed and visualized to draw appropriate conclusions.

In this exercise, you will train a neural network on the Heart failure clinical records Data Set to predict if the patient will die or live in the followed-up period. We are going to model this problem as a binary classification problem which detects whether the patient will live or die in the follow-up periodm, based on the other features. This example is built using custom training, but we could've used the pre-made keras fucnctions for training like model.comple and model.fit. The core of custom training is using the model to calculate the logits on specific set of inputs and compute the loss(in this case binary crossentropy) by comparing the predicted outputs to the true outputs. We then update the trainable weights using the optimizer algorithm chosen. The optimizer algorithm requires our computed loss and partial derivatives of loss with respect to each of the trainable weights to make updates to the same. We use gradient tape to calculate the gradients and then update the model trainable weights using the optimizer.

No description available

No description available

Classification Machine Learning task - "Heart failure clinical records Data Set" (UCI DATA SET)

Using L2X (Learning to Explain) algorithm to train a model on clinical heart failure data set.

Developed a model to predict a person's Death due to heart failure using some essential features ,using the heart-failure-clinical-data set easily available on Kaggle.

Heart failure clinical records Data Set contains the medical records of 299 patients who had heart failure. The dataset contains 11 clinical features (some of them are binary, others are numerical), the follow-up period and the label DEATH_EVENT that indicates whether or not the patient has died.

The dataset for exploration, modeling, and interpretability, explainability is called "Heart failure clinical records Data Set" to be found at the UCI (University California Irvine) Machine Learning Repository.

A machine learning model built for predicting heart failure from a set of clinical data. Logistic regression, random forest and K nearest neighbor models were trailed, with mean accuracies between 84-86%. .

It is a ML based SOLUTION(algorithms used: Decision tree, Random forest, XG-Boost, SVM) for heart failure prediction that uses the clinical dataset from kaggle to analyze and work up on algo based on the provided data set

In the second part of the test, we will be using heart failure clinical record data set. We will be using artificial neural network and Bidirectional LSTM for prediction whether the patient survives or not and then we will be going forward with the counterfactual part of the implementation where we will be making our model explainable by in the case prediction made is that patient dies than a counter sequence of input is generated from which we can change the prediction to positive one (that he survives). The method we will be using for the counterfactual part will be 1 NN Baseline Method.

Data Set Information: A detailed description of the dataset can be found in the Dataset section of the following paper: Davide Chicco, Giuseppe Jurman: "Machine learning can predict survival of patients with heart failure from serum creatinine and ejection fraction alone". BMC Medical Informatics and Decision Making 20, 16 (2020). Attribute Information: Thirteen (13) clinical features: age: age of the patient (years) anaemia: decrease of red blood cells or hemoglobin (boolean) high blood pressure: if the patient has hypertension (boolean) creatinine phosphokinase (CPK): level of the CPK enzyme in the blood (mcg/L) diabetes: if the patient has diabetes (boolean) ejection fraction: percentage of blood leaving the heart at each contraction (percentage) platelets: platelets in the blood (kiloplatelets/mL) sex: woman or man (binary) serum creatinine: level of serum creatinine in the blood (mg/dL) serum sodium: level of serum sodium in the blood (mEq/L) smoking: if the patient smokes or not (boolean) time: follow-up period (days) [target] death event: if the patient deceased during the follow-up period (boolean)