Analysing Channel Surfing Behaviour of IPTV Subscribers Using Machine Learning Models

Abstract

The Internet's pervasiveness has resulted in major shifts in the television ecosphere, where IPTV subscribers are now able to stream their favourite TV channels without having to consider time or location. Channel surfing is the practice of quickly scanning through various television channels in search of something interesting to watch. Due to the large number of TV channels available to IPTV subscribers, these subscribers may have difficulty matching their channel interests. This study aims to use machine learning models to analyze IPTV subscribers' channel surfing behaviours and predict contributing factors that lead to the rapid change of IPTV channels. Logitboost was benchmarked with six machine learning models in analyzing IPTV subscribers' channel surfing behaviour. Consequently, eight well-known performance evaluation metrics were used to compare the effectiveness of the machine learning models. The result presented shows that Logitboost outperformed the other six machine learning models. Consequently, the study identified four significant features that contribute to the channel surfing behaviour of IPTV subscribers which includes gender, peak hour, age, and genre. The findings show that over 40% of channel switching occurrences occur in less than 10 seconds, indicative that user attentiveness is very unpredictable. The study further discovered significant gender variations in channel genre viewing behaviours during peak hours.