Should I use LSTM or Dense or Some other Keras layer for time series data input layer?

Question

I have a tabular data of [10000 rows x 35 columns], each row depicting daily observations of 5 departments each providing 7 features. Thus, total 35 features are there as observation per day wrt the growth output (1 numeric value) for every 10000 days.

These 35 features may have a linear or non-linear mapping with response output on daily basis. I want to map it via NN and came across LSTM (for time series) and Dense Keras layers.

If I try to map it via the LSTM layer, the input shape will seek timesteps and features as parameters and a 3D array. As per use cases I observed of LSTM, it means that I convert this data like, create a list of list of "first 50 or XYZ timesteps of 35 feature" followed by +1 timesteps array. This is like providing 50 or XYZ days data to get the result of the 50th day and then repeating the same for 1-51 days, 2-52 days, 3-53 days, and so on for the whole data.

model.add(LSTM(25, return_sequences=True, input_shape = (50,35)))
model.add(LSTM(25, return_sequences=False))
model.add(Dense(10))
model.add(Dense(1))

Is this the best way to find relationships among time-series data or using simple layers like Dense will work or are there some better options?