Inverse-Distance Weighted K- Nearest Neighbor for Raw and Scaled Data set in Human Detection using Odour

Weights and Scaling of the input data set a pillar factors affecting the increase of efficiency of the AI model used in classification. In this paper, we investigate the effect of Inverse - Weight and three different scaling methods on the detection of humans from Odour. The Standardization, Normalization, and Logarithmic scalar have been applied on the input data set which is 15 scalar values each represent a gas value, collectively can identify the human from Odour. Each scaling method has been applied to the K-Nearest Neighbor method to detect classification accuracy. K-Nearest Neighbor with a variety of K values and functions have been tested. Non-scaled Raw data have been applied to detect the best AI Model and bring the best results from a raw data set. Human Odour has been collected from the body's armpits sweat with a gas chromatography device. 15 individuals of both genders and a variety of ages were invited to collect the required data set. Results show that scaling does not affect human detection, while the Inverse - Distance Weight was the crucial key that solved the problem and brought 100% accuracy.