Comparison of two feature extraction approaches from a large-scale bio-signal dataset

To understand how behavior state is represented in the central nervous system, a large-scale functional screen of AN movement encoding, brain targeting, and motor system patterning was performed in the adult fly, Drosophila melanogaster. However, based on different feature extraction methods, a different message is concluded and misinterpreted the dataset. This project compares the results between two approaches, which are linear regression and a resampled voting mechanism of statistical analysis. Regularized linear regression tends to generate cleaner results by restricting the explained variance to a single variable, while the voting statistical approach retains the possibility of showing multiple variables contributing to the correlation of the signals. Therefore, linear regression may mislead the understanding of the encoding pattern of behavior, neglecting the multi-encoding types of neurons. The voting statistical approach retain the nature better but the summary result are more noisy. In conclusion, we suggested to combine two approaches to guide yourself when navigating the dataset. With this analysis, we discovered that ascending populations convey high-level behavioral state which indicates that the spinal cord hosts the local computation for behavior control.