Data Analytics to Predict the Survivability of a Lost Person with Dementia Using R

This paper presents the results of analyzing cases of lost persons with dementia recorded in the International Search and Rescue Incident Database (ISRID). Linear regression, logistic regression and classification models were applied to determine the best method for predicting the survivability of such a lost person. The goal of this study is to understand the behavior of the lost person and determine what significant variables enhance their survivability. An informed cleaning process involving both manual and `R'-automated approach to scrub and augment the data--adding any missing values in the dataset. Linear regression is proposed to acquire the correlation among the numeric values in the database. There was no significant correlation among the independent variables. However, the data indicated that the wanderer tend to be found closer to the where they left or were last seen. They tend to go further in flat terrains and male wanderer tend to travel further than female wonderers Then, logistic regression was used to investigate the survivability using three classification models. As the number of persons found alive greatly exceed those deceased, the classifiers tends to be biased towards the majority class. Therefore, our focus was on finding the best model to provide the most accurate prediction. The three methods used were the Random Forest Classification Method, Rose Sampling and Synthetic Minority Oversampling Technique (SMOTE). The later, SMOTE, proved to be the best method, the Rose sampling produced errors and the random forest classification method produced over-fitting result. The outcome of this work will help inform the design of an algorithm and/or framework to be used for searches intended to be conducted by Unmanned Ariel Vehicles (UAVs). The outcome will also inform the design of use cases that will be used in various test environments.