Complex Data Imputation by Auto-Encoders and Convolutional Neural Networks—A Case Study on Genome Gap-Filling

• DOI: 10.3390/computers9020037
• Publication Date: 2020-05-11T16:26:30Z
• Authors: Luca Cappelletti; Tommaso Fontana; Guido Walter Di D...; Lorenzo Di Tucci; Elena Casiraghi; Giorgio Valentini
• Keywords: Imputation (statistics)

Missing data imputation has been a hot topic in the past decade, and many state-of-the-art works have presented to propose novel, interesting solutions that applied variety of fields. In successful results achieved by deep learning techniques opened way their application for solving difficult problems where human skill is not able provide reliable solution. Not surprisingly, some learners, mainly exploiting encoder-decoder architectures, also designed task missing imputation. However, most proposed tackle “complex data”, high dimensional belonging datasets with huge cardinality describing complex problems. Precisely, they often need critical parameters be manually set or exploit architecture and/or training phases make computational load impracticable. this paper, after clustering into three broad categories, we briefly review representative methods then describe our proposals, which specifically handle data. Comparative tests on genome sequences show imputers outperform KNN-imputation method when filling gaps sequences.