The rise of IoT and cloud technologies in both residential and enterprise networks has contributed significantly to the expansion of botnets. Considering this aspect, it is likely that such IoT bots can launch black-box data poisoning attacks on wireless networks which use AI/ML-based network optimizations. Especially when such network optimizations are used in closed loop, data poisoning attacks leading to corruption of training data can be disastrous for the wireless networks. Hence, studying the effects of different flavours of data poisoning attacks on various ML algorithms in wireless networks is an important aspect from an ML security point of view. In this paper, we present experiments performed to assess the impacts of data poisoning on various ML algorithms. The dataset utilized in these experiments is derived from simulated environments, with both offline training and validation procedures implemented under conditions of poisoned and non-poisoned data. The dataset includes key performance indicators (KPIs) pertinent to latency and cell load and is publicly available on GitHub. In our experiments, poisoning of the label and poisoning of feature set is carried out for the cell load prediction use-case. In addition, these experiments facilitate the analysis of the impacts of data poisoning across varying sizes of the training dataset. Our experimental results indicate that data injection attacks present a substantial threat to AI/ML systems, requiring increased attention and countermeasures. Moreover, increasing the training dataset size appears to be a viable strategy to mitigate the adverse impacts of various data poisoning attacks.