1. Testing Protocols -Good scientific protocols and test plan. The results of your scans and data collections will only ever be as good as the data and scanning you enter and create. Therefore, a solid testing protocol, scientific standards and organization are critical to your success.
2. Data - as much as possible and as varied as possible. This point is critical for the success of your collections. By data we mean both spectrum and meta-data (attributes). The spectrum should be scanned properly and the meta-data should be obtained from a reliable source (such as an external lab or a different measurement device, etc.).
3. Scrubbing your data and selecting the appropriate pre-processing. "Processed" is mainly useful for estimation models. "Normalized" is useful mainly for classification models. Sometimes use of both "processed" and "normalized" is needed. You can also use the Expert mode, to create your own pre-processing method.
(Relevant Product: SCiO DevKit)