PurposeThe purpose of this paper is to employ data mining as a new diagnosing scheme for investigating void formation to the thermal pad in quad flat non‐lead (QFN) assembly. Occurrences of voiding in various scenarios of component design, materials selection and manufacturing process are analyzed.Design/methodology/approachThis research investigates the process yield of a PCB assembly for a handheld device in the electronics manufacturing industry using the chi‐square automatic interaction detection (CHAID) algorithm and chi‐square test. Practical data generated by an X‐ray apparatus from the shop floor are collected. The critical attributes to the void formation (in the solder joint) of the QFN component are identified.FindingsStocking the PCB material beyond ten days may increase the level of voiding by 1%. Using PCB provided by vendor U helps decrease the level of voiding by 1.6%. Stocking the component material above 43 days may increase the level of voiding by 1.9%. In addition, reflow soldering profile with time above liquid (TAL) less than or equal to 62 sec and with peak temperature higher than or equal to 241°C generate less voids. Finally, the via‐in‐pad design causes a concave geometry on the surface of thermal pad which contributes to the voiding formation. The amount of voiding can be further diminished by plugging the via with plated copper.Originality/valueThis research implements CHAID that extracts useful knowledge from a huge amount of manufacturing data in order to realize the complex interaction effects through automated analysis. The extent of voiding in the samples using the optimal process suggested through CHAID algorithm can be reduced from 16% to 10.2%.